Validation of Version 5.1 MODIS Aerosol Optical Depth (Deep Blue ...

1 downloads 0 Views 2MB Size Report
AOD values over water bodies (Michael King, personal communication). A small region of high AOD is observed over the Gulf of Khambhat, adjacent to Surat, ...
ARTICLE IN PRESS Aerosol and Air Quality Research, x: 1–11, xxxx Copyright © Taiwan Association for Aerosol Research ISSN: 1680-8584 print / 2071-1409 online doi: 10.4209/aaqr.2014.01.0004 

Validation of Version 5.1 MODIS Aerosol Optical Depth (Deep Blue Algorithm and Dark Target Approach) over a Semi-Arid Location in Western India Amit Misra1*, Achuthan Jayaraman2, Dilip Ganguly3 1

Indian Institute of Technology Kanpur, Kanpur, India National Atmospheric Research Laboratory, Gadanki, India 3 Indian Institute of Technology Delhi, New Delhi, India 2

ABSTRACT We have examined the Moderate Resolution Imaging Spectroradiometer (MODIS) Collection 5.1 Level 2 Aerosol Optical Depth (AOD) from Deep Blue algorithm and Dark Target approach over Ahmedabad, India for 2002 to 2005. Deep Blue algorithm is observed to be able to retrieve AOD over the Rann of Kuchchh, a region of high surface reflectance which poses difficulty for the Dark Target approach. Microtops sunphotometer measured AOD was used to validate the MODIS AOD from the two algorithms. MODIS data from both Terra and Aqua platforms were used, and the comparison was done at 470, 550, and 660 nm wavelengths. The MODIS (Dark Target) - Microtops correlation showed an improvement in all correlation parameters over that from MODIS Collection 5 at our study region. Considering the overall dataset, MODIS (Dark Target) - Microtops showed a stronger correlation (R2550 = 0.55 for Terra, R2550 = 0.69 for Aqua) than MODIS (Deep Blue) - Microtops (R2550 = 0.43 for Terra, R2550 = 0.50 for Aqua). A diurnal dependence of the error due to improper aerosol model assumption was noted with the slopes of MODIS-Microtops AOD correlation being lower for Terra than for Aqua. This feature was observed to affect the MODIS-Microtops AOD correlation for Dark Target approach more than the Deep Blue algorithm. Deep Blue derived AOD and Dark Target AOD from Aqua MODIS were found to be consistent, with the correlation between the AODs from the two algorithms being better for Aqua than Terra for all seasons, except monsoon. No dependence was observed of the correlation between MODIS-Microtops AOD on surface reflectance. Our recommendation is to use the Deep Blue AOD over the Rann of Kuchchh region and Dark Target AOD over other regions of Gujarat. Keywords: MODIS; AOD retrieval; Dark target approach; Deep blue algorithm; Validation.

INTRODUCTION The role of aerosols in the Earth climate system is a topic of ongoing research. This is on account of the role aerosols play in Earth atmosphere by affecting the incoming solar radiation by scattering and absorption (Haywood and Ramaswamy, 1998; Hatzianastassiou et al., 2007). In addition, aerosols affect the cloud microphysical properties and their lifetimes (IPCC Fourth Assessment Report: Climate Change, 2007). An assessment of aerosol amount and their properties is a pre-requisite to any study on their effect on Earth climate. However, the development of an aerosol climatology is challenging due to diversity in aerosol types and their sources along with the short residence time of these particles in the atmosphere which result in large spatial and temporal heterogeneity in the distribution of aerosols (King et al., 1999).

*

Corresponding author. E-mail address: [email protected]

It is in this context that the application of satellite remote sensing for the measurement and characterization of aerosols gains importance. Because of their large swath and routine unmanned monitoring, satellites can provide a detailed account of aerosol features on a much larger space- and time- scale than is possible from conventional means. This has further application in the study of transport of aerosols and the impact of non- local sources on the climatology of any region. Application of satellites for aerosol monitoring over oceans has been accomplished with sufficient accuracy in the past (Tanre et al., 1997). However, satellite based aerosol remote sensing over land surfaces continues to be a formidable task. The uncertainty in the aerosol retrieval over land mainly stems from the uncertainties related to aerosol model selection and surface reflectance (Kaufman et al., 1997a). Cloud contamination is an additional source of error in satellite based aerosol retrieval. The aerosol models are largely based on the ground based sunphotometers, and their appropriateness depends on the number of such ground based observation stations. The latter is the main factor

ARTICLE IN PRESS 2

Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

governing the accuracy of retrieved AOD with respect to aerosol model. Though the initial versions of the MODIS aerosol product used the aerosol models based on the best available data at the time, the revised algorithm version and updated product utilize the aerosol climatology obtained by the AERONET sunphotometer network (Levy et al., 2007a). Since the whole globe is not covered by this network, some uncertainty is still left, e.g., over the regions where large heterogeneities are present. However, this is a limitation related to the available ground based database, and cannot be considered a limitation of the aerosol retrieval algorithm. Residual cloud contamination is a source of error in the satellite based remote sensing in all the disciplines - land, ocean, or atmosphere study. Further, within the ambit of atmosphere studies, the biggest source of error in the measurements from satellite based passive remote sensing arises from surface reflectance. This is specially the case over land as compared to over ocean since in the latter case, the ocean surface provides a dark background against which atmospheric aerosols are mapped. Kaufman et al. (1997a) mention the error in retrieved AOD to be 10 times the error made in the estimation of surface reflectance. The errors are higher when the surface is highly reflecting such as deserts, snow, and ice sheets. Further, since the resulting aerosol product involves averaging of several pixels, any sharp heterogeneity at the boundary of such surfaces result in erroneous results for neighbouring locations as well. As a result, the prime concern in the retrieval of aerosol properties from satellite measured radiance is to remove the contribution from surface reflectance. The Dark Target approach (DT) currently employed by the MODIS aerosol group tried to account for surface reflectance in visible channels based on the observations in mid-IR channels (Kaufman et al., 1997a; Remer et al., 2005). After initial assessment of the retrieved aerosol properties (Chu et al., 2002; Ichoku et al., 2002; Levy et al., 2005) and sensitivity studies on individual parameters (Gatebe et al., 2001; Remer et al., 2001), the surface reflectance parameterization was modified (Levy et al., 2007b). Alternative algorithms have also been proposed by several groups for satellite based aerosol remote sensing. This paper examines the Deep Blue algorithm (DB) (Hsu et al., 2004) which uses the observations in the blue channel, avoiding the usage of red wavelength where the surface reflectance values are relatively large. More discussion about this method is given later in the paper. It must be noted that the appropriateness of any method lies in the efficiency with which it is able to retrieve the aerosol properties with least possible assumptions and minimum errors. Therefore, after the completion of aerosol parameter retrieval and the release of corresponding products, an exhaustive validation of the data against ground truth data is needed. These ground based observations - mostly from sunphotometers retrieving aerosol optical properties from the measurement of direct solar radiation - are free from surface reflectance related error, and nearly free from cloud contamination. Thus, these data provide the benchmark to test the efficacy of the retrieval algorithm for aerosol remote sensing from space. They also show the direction for further modifications in the developed algorithm. For

example, a detailed validation of an initial version of an algorithm may provide an insight into the cases where the performance of the procedure is satisfactory and the cases where further improvement is needed. Such insight provides the necessary guideline and roadmap for a better procedure. Several validation efforts by several groups (e.g., Chu et al., 2002; Ichoku et al., 2002; Levy et al., 2005, 2010) have provided the impetus for update and upgrade of the MODIS aerosol retrieval algorithm from time to time. The initial algorithm proposed by Kaufman et al. (1997a) has been subsequently modified, and the latest changes are described by Levy et al. (2007b) and Hsu et al. (2004) which retrieve aerosol properties by following two different approaches. While Levy et al. (2007b) focus on the modification and upgrade of the earlier method by Kaufman et al. (1997a) and Remer et al. (2005), the procedure followed by Hsu et al. (2004) is a completely different and alternative approach to address the problem. For the sake of clarity, we will use the notation ‘DT’ for the latest version of the Dark Target approach (Levy et al., 2007b), and ‘DB’ for the Deep Blue algorithm (Hsu et al., 2004). The present paper reports validation of the MODIS version 5.1 DB aerosol product from both Terra and Aqua at Ahmedabad, India, for 2002–2005. In addition, salient features from validation results of DT version 5.1 from both Terra and Aqua are also included and compared with DB validation results. The latter extends the study by Misra et al. (2008) to the version 5.1 product, and also includes the data from Terra platform. Incorporating the data from both Terra and Aqua provides indication of any diurnal dependence of the comparison results possibly arising due to variation in aerosol properties, surface characteristics, and solar and satellite geometry. While the DT approach has been validated previously over India using earlier versions of the MODIS AOD product (Tripathi et al., 2005; Jethva et al., 2007a, b; Prasad and Singh, 2007; Misra et al., 2008; Choudhry et al., 2012), the current study performs a validation of the latest version (MODIS collection 5.1). Thus, this paper provides an update to the results of previous MODIS AOD studies. In addition, this study provides one of the first attempts to validate the DB algorithm over India. Section 2 outlines the MODIS aerosol retrieval algorithms, section 3 describes the study location and local meteorology. Results of data analysis and validation are mentioned in section 4, and conclusions in section 5. THE MODIS AOD RETRIEVAL ALGORITHMS The Dark Target (DT) Approach Since the launch of MODIS intruments in 1999 and 2002 onboard Terra and Aqua satellites, respectively, several revisions have been made to the aerosol retrieval algorithm, though the underlying principle remains the same (Kaufman et al., 1997a; Remer et al., 2005). The surface reflectance at 470 and 660 nm is first estimated from empirical relations between surface reflectance at visible and mid- IR wavelengths. The aerosol type is decided based on the ratio of path radiance at 470 and 660 nm and the geographical location of the region of study. Finally, after cloud screening

ARTICLE IN PRESS Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

3

and correcting for gas absorption, satellite measured radiances are compared to the simulated radiances for different solar and satellite geometry, aerosol optical depth, and surface reflectances to find the value of aerosol optical depth using the look-up table approach. This technique is the DT approach and provides AOD assessments for surface reflectance values up to 0.15. The reflectance for most deserts and snow covered areas are higher than 0.15, so the retrieval algorithm fails for such cases. Further, the underlying principle of transparency of aerosols at mid-IR wavelength, which is the backbone of this method, breaks down when coarse size particles are present (Kaufman et al., 1997a, b). The update to the retrieval algorithm by Levy et al. (2007b) is a major overhaul over the initial version. The updated product takes polarization into account while performing the radiative transfer calculation for look-up table generation. Additional changes include angular and seasonal dependence of surface reflectance ratios at visible and mid-IR wavelengths, and usage of the latest aerosol climatology from the worldwide AERONET sunphotometer network (Dubovik et al., 2002; Levy et al., 2007a). Performance of the Collection 4 and 5 of the MODIS DT aerosol product over India has been thoroughly tested by several groups (Tripathi et al., 2005; Jethva et al., 2007a, b; Prasad and Singh, 2007; Misra et al., 2008), most recently by Choudhry et al. (2012).

follows a similar approach but deriving the aerosol optical depth and the fraction of dust and smoke aerosol in the composite at 412 and 490 nm. This technique is expected to retrieve aerosol optical depth values when surface reflectance at 670 nm is up to 0.3. AOD retrieval over surfaces with reflectance as high as 0.4 at 670 nm can also be accomplished provided AOD is greater than 0.7. The major sources of error include surface reflectivity, aerosol vertical profile, and uncertainty related to the aerosol shape. The initial validation of the retrieved aerosol optical depth with AERONET sunphotometer data at Ilorin (Nigeria) and Solar Village (Saudi Arabia) showed promising scope of the algorithm with the calculated AOD values within 20–30% of the AERONET observed AOD values (Hsu et al., 2004). The changes in the collection version 5.1 from the previous version include better surface bidirectional reflectance distribution function (BRDF) and terrain effects accounting, and improved cloud screening. Several errors in the collection version 5 have also been rectified in the updated product (MODIS-Atmosphere Collection 051 Changes Version 01 document). The DB aerosol product is being used for various studies on aerosol climatology and transport (e.g., Gautam et al., 2010, 2011).

The Deep Blue (DB) Algorithm The DB algorithm (Hsu et al., 2004) was developed considering difficulty in aerosol retrieval by space based instruments over arid, semi arid, and urban areas. The underlying principle is the low surface reflection of such surfaces at blue as compared to the red portion of the visible spectrum. DB algorithm exploits this property by using the surface reflectance in blue region of the visible spectrum for the retrieval process. The major steps of the retrieval process are as follows: Assuming the underlying surface to be Lambertian and homogeneous, look-up tables for satellite level radiances are generated using a polarized radiative transfer model (Dave, 1970). Two different procedures are followed for cloud screening of the radiance data at the preprocessing stage: 1) reflectance at 412 nm to separate clear and cloudy pixels, and 2) aerosol index evaluation using 412 and 490 nm to distinguish thick dust from cirrus clouds. Based on geolocation, a surface reflectance database of surface reflectivity is generated at 0.1° latitude × 0.1° longitude grid at 412, 490, and 670 nm using minimum reflectivity technique. Finally, the reflectance measured by the satellite is compared with the corresponding values in the look-up table to derive the values of aerosol optical depth and single scattering albedo using the maximum likelihood method. Depending on the aerosol loading, two different approaches are followed for final aerosol retrieval: for higher aerosol loading (τ > 0.7), radiances at 412, 490, and 670 nm are used, whereas for low and moderate aerosol loading (τ < 0.7), only two wavelengths viz., 412 and 490 nm are used. This is due to the reduced sensitivity of radiance at 670 nm to aerosol scattering because of high surface reflectance at this wavelength. For mixed type aerosols, the method

The study region is the western Indian state of Gujarat (68–75°E, 19–25°N) (Fig. 1(a)). The state is heterogeneous in aerosol types and surface features. Gujarat is bound on the west and south-west by the Arabian sea, and Thar desert in north. The eastern and south-eastern region is largely dominated by continental type aerosols. The north-west region of Gujarat, called the Rann of Kuchchh, is a broad, low-lying area of salt deposits. The high surface reflectance poses problem for satellite based remote sensing of aerosols over this region. For example, the earlier MODIS Collection C004 aerosol product gave missing values for this area for most of the months. As it has been mentioned in the previous two sections, with the updated MODIS aerosol product and also the alternative product implementing the DB algorithm, considerable hope has been generated for getting the aerosol optical depth values over this region. The DB algorithm is especially suited to aerosol remote sensing over such ‘bright reflecting surfaces’. This also highlights another vantage point for satellite remote sensing, because this region is a remote location and aerosol monitoring from in-situ ground based measurements is not viable. Thus, the heterogeneity in aerosol types and surface features make Gujarat a good test-bed where the aerosol maps from the two approaches can be juxtaposed to infer their comparative advantage under diverse conditions. This is done in the next section where the aerosol maps over Gujarat from the two MODIS aerosol products are examined. The location chosen for validation of MODIS aerosol products is Ahmedabad (72.5°E, 23.03°N) in Gujarat. The site is semi-arid and is dominated by mineral dust aerosols. The meteorology of Ahmedabad, which provides the background for interpretation of the validation results, has been detailed in Ganguly et al. (2006) and Misra et al. (2008),

SITE DESCRIPTION

ARTICLE IN PRESS 4

Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

Fig. 1. Annual mean AOD climatology at 550 nm over Gujarat from MODIS for the period 2002–2005. Panel (a) shows a map of the study region including locations mentioned in the text. Background image in panel (a) is taken from Google Earth. Panels (b), (c), and (d) are based on Terra MODIS data, whereas panels (e), (f), and (g) are based on Aqua MODIS data. Panels (b) and (e) correspond to DB algorithm, panels (c) and (f) correspond to DT approach, and panels (d) and (g) depict the difference between DB and DT derived AOD over the Gujarat region. and only the main points are mentioned here. For more details, one should refer Ganguly et al. (2006) and Misra et al. (2008). On the basis of meteorological parameters like temperature, wind speed, wind direction and rainfall, the year is divided into four seasons viz, Dry (December to March), Pre-Monsoon (April and May), Monsoon (June to September) and Post-Monsoon (October and November). RESULTS AND DISCUSSION Comparison of DT and DB Derived Aerosol Climatology Fig. 1 shows the plot of AOD at 550 nm over Gujarat from DT and DB aerosol products, and the difference between DB and DT derived AOD (i.e., AODDB - AODDT). The AODs shown are annual mean climatology for the period 2002–2005 derived using the MODIS Level 3 monthly mean data products. The top panel is for Terra MODIS whereas the bottom panel is for Aqua MODIS. The most important feature observed from the data is that DB fills in the missing values present in the DT derived AOD maps, especially the Rann of Kuchchh. Highest values of AOD are observed over the Rann of Kuchchh in both the DB and DT derived maps. This is a region with large spread of salt lakes and has high surface reflectance values. AOD retrieval over this region is a challenging task due to

the high surface reflectance. DT derived AOD maps usually show missing values over this region, and its retrieved values are usually higher than DB derived values. While DT derived maps are able to distinguish the higher AOD values over the industrial areas in south-east Gujarat compared to other adjoining areas, this contrast is weakly captured in DB derived maps. AOD values are seen to be low over northeast and east Gujarat in both DB and DT maps, and over Arabian Sea in DT maps. At present, DB does not retrieve AOD values over water bodies (Michael King, personal communication). A small region of high AOD is observed over the Gulf of Khambhat, adjacent to Surat, which is conspicuous in the DT derived AOD maps from both Terra and Aqua. Though this feature is not very prominent in the DB derived maps, yet the AOD over the Gulf of Khambhat in DB maps is higher than the surrounding areas. Several interesting features of this comparison can be noted from the spatial maps showing the differences in AOD over Gujarat derived using DB and DT techniques. It is observed that DB derived AOD are higher than DT derived AOD over north Gujarat, while over south Gujarat, DT derived AOD are higher than DB derived AOD. Over the central Gujarat, i.e., the boundary of north and south Gujarat, DT and DB derived AOD values are nearly similar. Overall DT climatology is similar for Aqua and Terra. It may be noted

ARTICLE IN PRESS Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

from Fig. 1 that the difference between DB and DT derived AODs from the same sensor (Aqua or Terra) is larger compared to differences between the AOD values derived from different sensors using the same algorithm. Overall, DB derived AOD climatology along coastal Gujarat, southeast Gujarat, and the Rann of Kuchchh is similar from both Aqua and Terra. However, over north and central Gujarat, Aqua AOD values are slightly higher than those from Terra. Fig. 2 shows the seasonal mean AOD climatology over Gujarat for the period 2002–2005 derived using DB and DT versions of Terra MODIS. Large variation in AOD with season is observed over the Gujarat region. Lowest AOD is observed during post-monsoon and dry seasons, and the difference between DB and DT AOD are also lowest during these seasons. Highest AOD is observed during monsoon, followed by pre-monsoon season. Differences between the

5

two algorithms is also large during these seasons, with DB derived AOD greater than DT derived AOD over north Gujarat, and DT derived AOD greater than DB derived AOD over south Gujarat. The difference between DB and DT derived AODs is found to be dependent on the value of AOD itself. This difference is large during pre-monsoon and monsoon seasons when AOD values are high, while the difference is low during post-monsoon and dry seasons when AOD values are low. In the present work, we have illustrated the differences in annual mean and seasonal mean AOD distribution over the Gujarat region derived from the two approaches used by MODIS. Based on limited ground validation that we could make, as discussed in the next section, our recommendation to the MODIS user community is to use the DB derived AOD over the Rann of Kuchchh region and DT derived AOD over the rest of Gujarat.

Fig. 2. Seasonal mean AOD climatology at 550 nm over Gujarat from Terra MODIS for the period 2002–2005. Panels (a), (b), and (c) are for dry season; panels (d), (e), and (f) are for pre-monsoon season; panels (g), (h), and (i) are for monsoon season; and panels (j), (k), and (l) are for post-monsoon season. Panels (a), (d), (g), and (j) correspond to DB algorithm; panels (b), (e), (h), and (k) correspond to DT approach; and panels (c), (f), (i), and (l) give the difference between DB and DT derived AOD.

ARTICLE IN PRESS 6

Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

Ground Validation Fig. 3 shows the time series of monthly average AOD, calculated from Level 2 MODIS data, at Ahmedabad during 2002 to 2005 from DB and DT at 550 nm, and surface reflectance at 550 nm from Terra. In general, surface reflectance is observed to be high during June, July, August, and September months, with values greater than 0.3. For other months, its value is in general less than 0.2. Though surface reflectance is expected to remain low during monsoon months, still rainfall is not regular throughout the season. There are prolonged dry spells between intermittent rainfalls. However, the role of residual cloud contamination cannot be ruled out. DB AOD is lowest during September 2005 (AOD = 0.043), and highest during September 2003 (AOD = 0.92). DT AOD is lowest during February 2004 (AOD = 0.15), and highest during September 2003 (AOD = 0.55). For September 2003, May 2004, and April 2005, DB AOD is higher than DT AOD, whereas for all other cases, DT AOD is either higher than or equal to DB AOD. For June 2003, DT AOD is equal to DB AOD. We have attempted the validation of the MODIS derived Level 2 AOD (version 5.1) from DB algorithm and DT approach using ground based sunphotometer measurements. The Microtops sunphotometer is used for ground based AOD measurement (Morys et al., 2001). This instrument derives the columnar aerosol optical depth by measuring the attenuation in the direct solar radiation. The uncertainty in measurement of AOD by Microtops sunphotometer is less than 0.03 (Ganguly et al., 2006). Besides studying the overall comparison, the datasets are also compared for different years and seasons. We have considered the MODIS Level 2 data from Terra and Aqua platforms separately in order to explore any diurnal variation of the correlation parameters. We have compared the AODs at 470 nm, 550 nm, and 660 nm. The equation of linear correlation reveals information on the factors affecting the correlation. The slope (m) of the equation denotes the error due to incorrect aerosol model assumption, whereas the intercept (c) shows improper surface reflectance parameterisation. In the ideal case of perfect match between the space-based and ground-based AOD measurements, m = 1, and c = 0. The criteria for collocation of MODIS and Microtops AOD are the same as described in Misra et al. (2008). The Angstrom fitting of Microtops AOD to find ground based AOD at MODIS wavelengths is also the same as in Misra et al. (2008). AOD data DT and DB SDS from Level 2 MODIS aerosol product are examined with the help of Microtops AOD collocated in space, time, and wavelength. In addition, MODIS surface reflectance data is used to explore any dependence of correlation between MODIS and Microtops AOD on surface reflectance. Fig. 4 shows the overall correlation between MODIS and Microtops derived AOD for Terra (left column) and Aqua (right column) at 470 (top row), 550 (middle row), and 660 nm (bottom row). Data points and equation depicted in blue color correspond to DB algorithm, and those in black color correspond to DT approach. Higher correlation (R2) is observed for Aqua than Terra (at 550 nm, R2Aqua = 0.50,

ARTICLE IN PRESS Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

7

Fig. 4. Overall correlation between MODIS and Microtops derived AOD for Terra (left column) and Aqua (right column) at 470 (top row), 550 (middle row), and 660 nm (bottom row). Blue color and ‘X’ sign corresponds to Deep Blue algorithm, and black color and ‘+’ sign corresponds to Dark Target approach. Number of data points are NTerra,DB = 213, NTerra,DT = 231, NAqua,DB = 184, NAqua,DT = 187. R2Terra = 0.43 for DB; R2Aqua = 0.69, R2Terra = 0.55 for DT), which increases with wavelength. This feature is more pronounced in DT derived correlation (e.g., for Aqua, R2470 = 0.62, R2550 = 0.69, R2660 = 0.75). Improvement in values of slopes is also observed with increasing wavelength (e.g., for DT - Microtops comparison from Aqua, m470 = 0.86, m550 = 0.91, m660 = 0.98), and slopes are found to be more closer to unity for DB algorithm (m550 = 0.96 for Aqua, m550 = 0.80 for Terra) than DT approach (m550 = 0.91 for Aqua, m550 = 0.59 for Terra). However, correlation (R2) is better for DT approach (R2550 = 0.55 for Terra, R2550 = 0.69 for Aqua) than DB algorithm (R2550 = 0.43 for Terra, R2550 = 0.50 for Aqua). This is the case for all wavelengths and both platforms. Considering overall correlation, intercepts are negative for all wavelengths and both platforms for DBMicrotops comparison (c550 = –0.08 for Aqua, c550 = –0.08 for Terra), implying slight overcorrection for surface

reflectance. However, for DT-Microtops comparison, the intercepts are positive for Terra (c550 = 0.09) and negative for Aqua (c550 = –0.05) cases. It is interesting to note that the difference in best fits for DB vs Microtops and DT vs Microtops is much larger for Terra than Aqua. This is mainly due to the fact that the underestimation is more for Terra than Aqua, and further, the difference is more for DT than DB, i.e., the degree of underestimation by Terra MODIS is more for DT than DB. Tables 1 and 2 give the parameters of correlation analysis (m, c, and R2) for different years and seasons. One interesting feature is noted in the slopes of seasonal correlation analysis for DT approach. It is noted that slopes are lower for 470 nm than 660 nm during dry (m470 = 0.35, m660 = 0.49) and post monsoon (m470 = 0.72, m660 = 0.78) seasons, whereas the slopes are lower for 660 nm than 470 nm during pre-monsoon season (m470 = 1.13, m660 = 1.08). As

 

Overall 2002 2003 2004 2005 Dry PreM Mon PosM

Case

m 0.56 0.52 0.46 0.59 1.16 0.35 1.13 0.36 0.72

m 0.79 0.21 1.05 0.44 1.38 0.51 1.76 1.01 0.54

470 nm c 0.11 0.15 0.15 0.11 –0.15 0.15 –0.04 0.28 0.01

470 nm c –0.09 0.13 –0.18 0.05 –0.35 0.02 –0.24 –0.15 –0.06

m 0.8 0.26 0.92 0.48 1.49 0.47 1.76 0.91 0.5

R2 0.43 0.11 0.69 0.24 0.62 0.29 0.74 0.68 0.32 m 0.89 0.35 0.91 0.55 1.6 0.45 1.79 0.92 0.52

660 nm c –0.1 0.06 –0.1 0 –0.33 0.02 –0.25 –0.12 –0.03 R2 0.49 0.16 0.73 0.31 0.73 0.29 0.83 0.7 0.32 m 0.91 0.93 0.77 0.62 1.36 0.88 1.5 0.86 0.71

470 nm c –0.07 –0.08 –0.01 0.03 –0.21 –0.06 –0.17 0.08 –0.11 R2 0.45 0.68 0.32 0.33 0.62 0.56 0.82 0.22 0.51 m 0.96 0.83 0.82 0.62 1.39 0.8 1.46 0.92 0.65

Aqua 550 nm c –0.08 –0.03 –0.04 0.02 –0.2 –0.03 –0.16 0.04 –0.07

R2 0.5 0.4 0.61 0.51 0.69 0.33 0.73 0.55 0.65

m 0.59 0.61 0.44 0.66 1.19 0.39 1.08 0.34 0.72

Terra 550 nm c 0.09 0.11 0.15 0.07 –0.14 0.13 –0.03 0.25 0.02 R2 0.55 0.52 0.59 0.59 0.79 0.34 0.78 0.52 0.69 m 0.69 0.72 0.49 0.74 1.19 0.49 1.08 0.39 0.78

660 nm c 0.06 0.07 0.12 0.05 –0.12 0.1 –0.04 0.21 0.01 R2 0.64 0.67 0.6 0.68 0.85 0.37 0.84 0.53 0.78

m 0.86 0.93 0.69 0.74 1.26 0.69 1.41 0.84 0.79

470 nm c –0.04 –0.02 0.01 0.02 –0.16 0.03 –0.2 0.06 –0.07

R2 0.62 0.83 0.51 0.66 0.75 0.62 0.8 0.64 0.75

m 0.91 0.85 0.71 0.79 1.29 0.74 1.33 0.79 0.82

Aqua 550 nm c –0.05 0.02 0.01 0 –0.16 0.02 –0.17 0.07 –0.06

Table 2. Parameters of MODIS (Dark Target) derived AOD vs. Microtops measured AOD correlation.

R2 0.4 0.09 0.7 0.2 0.53 0.3 0.67 0.68 0.32

Terra 550 nm c –0.08 0.1 –0.11 0.03 –0.35 0.03 –0.24 –0.11 –0.04

R2 0.69 0.85 0.53 0.7 0.84 0.62 0.84 0.64 0.77

R2 0.5 0.65 0.38 0.33 0.69 0.53 0.86 0.27 0.49

m 0.98 0.78 0.84 0.82 1.3 0.89 1.28 0.78 0.87

m 1.08 0.72 1.05 0.65 1.41 0.82 1.42 0.87 0.62

660 nm c –0.05 0.02 –0.02 0 –0.13 –0.01 –0.15 0.05 –0.05

660 nm c –0.11 –0.02 –0.11 0.01 –0.19 –0.04 –0.14 0.05 –0.04

R2 0.75 0.87 0.61 0.71 0.88 0.64 0.87 0.6 0.78

R2 0.59 0.58 0.53 0.36 0.76 0.54 0.88 0.24 0.46

8

Overall 2002 2003 2004 2005 Dry PreM Mon PosM

Case

Table 1. Parameters of MODIS (Deep Blue) derived AOD vs. Microtops measured AOD correlation.

ARTICLE IN PRESS

Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

ARTICLE IN PRESS

660 nm c –0.05 0 –0.05 –0.02 –0.05 –0.02 0.03 –0.09 –0.02 m 1.02 0.69 1.08 0.86 1.05 0.88 1.06 1.24 0.71 R2 0.82 0.71 0.75 0.67 0.92 0.8 0.96 0.68 0.67 R2 0.81 0.67 0.72 0.7 0.91 0.82 0.97 0.66 0.63

m 1.06 0.7 1.12 0.94 1.08 1.07 1.05 1.25 0.8

Aqua 550 nm c –0.05 0.01 –0.05 –0.03 –0.05 –0.05 0.04 –0.1 –0.03 R2 0.73 0.24 0.71 0.64 0.94 0.6 0.96 0.78 0.49

m 1.09 0.69 1.16 1 1.08 1.25 1.02 1.23 0.88

470 nm c –0.05 0.03 –0.05 –0.04 –0.05 –0.08 0.05 –0.1 –0.04 R2 0.73 0.24 0.76 0.63 0.94 0.64 0.96 0.82 0.48

m 1.33 0.57 1.8 0.9 1.47 0.84 1.51 2.45 0.69

660 nm c –0.16 0 –0.27 –0.07 –0.18 –0.05 –0.15 –0.55 –0.02 R2 0.72 0.24 0.8 0.61 0.94 0.67 0.95 0.85 0.46

m 1.43 0.57 2.11 0.94 1.49 1.05 1.51 2.76 0.73

Terra 550 nm c –0.19 0.01 –0.36 –0.07 –0.19 –0.09 –0.16 –0.69 –0.03 Overall 2002 2003 2004 2005 Dry PreM Mon PosM

m 1.52 0.57 2.34 0.97 1.47 1.26 1.48 2.94 0.75

470 nm c –0.23 0.02 –0.46 –0.08 –0.2 –0.14 –0.17 –0.83 –0.03 Case

Table 3. Parameters of Deep Blue derived AOD vs. Dark Target derived AOD correlation.

dry and post monsoon season are dominated by fine particles, and pre-monsoon season by coarse particles, the above observation implies an underestimation of the dominant aerosol species for these seasons. However, this feature is not reflected in the slopes of correlation analysis for monsoon season (m470 = 0.36, m660 = 0.39). Parameters of correlation between DB and DT derived AODs are presented in Table 3. Considering overall correlation, the correlation is seen to be better for Aqua (R2550 = 0.82) than Terra (R2550 = 0.73). For Aqua, the AOD values are nearly equal from both DT and DB, and the slope of correlation is nearly equal to unity (at 550 nm, mAqua = 1.06). For Terra derived values, the slope between DB and DT is much larger than unity (at 550 nm, mTerra = 1.43). These results present quantitatively the information conveyed in Fig. 3 that has been discussed in a previous paragraph. Thus, considering the correlation parameters for the overall comparison, the AOD derived by DB algorithm and DT approach are consistent for Aqua MODIS. The values of the correlation parameters vary with season. However, except for monsoon, the DB - DT comparison for Aqua MODIS is superior to Terra MODIS for all other seasons. We explored the possibility of dependence of the correlation between R2 of MODIS AOD vs Microtops AOD on surface reflectance. We evaluated the correlation between MODIS and Microtops AOD for individual months, and correlated the R2 values, thus obtained, with monthly averaged surface reflectance values. At 550 nm, the R2 value between surface reflectance and R2 (of MODIS-Microtops comparison) was 0.05 for DT, and 0.06 for DB. Excluding the cases with surface reflectance > 0.2 (to circumvent possible cloud contamination in surface reflectance data), reduces this value to R2 = 0.01 at 550 nm for both DT and DB. Thus we conclude that no correlation exists between these parameters, so that there is no dependence of MODIS-Microtops correlation on surface reflectance for our study. The comparison of DT and Microtops AODs (Table 2) show an improvement in all correlation parameters over the values obtained using the previous version of the DT algorithm (MODIS collection 5) over our study region (Misra et al., 2008). The slope, intercept, and R2 were 0.69, 0.03, 0.61, respectively, at 470 nm, and 0.80, 0.003, 0.69, respectively, at 660 nm, in Aqua MODIS-Microtops comparison for Collection 5 (Misra et al., 2008). These parameters have improved in the present study with slope, intercept, and R2 as 0.86, -0.04, 0.62, respectively, at 470 nm, and 0.98, –0.05, 0.75, respectively, at 660 nm in Aqua MODIS-Microtops comparison for Collection 5.1. Though this is the first validation attempt of the DB algorithm over our study region, the algorithm has been evaluated elsewhere. Li et al. (2012) have compared the MODIS aerosol product from Collection 4 and 5 over north-west China. They found that 73.3% of the DB retrievals fall within the error range of 30%. Shi et al. (2013) have evaluated the DB aerosol product (Collection 5.1) focusing on Arabian Peninsula and north Africa. They identified the accuracy of DB algorithm derived AOD to be dependent on aerosol microphysics and surface albedo. Xie et al. (2011)

R2 0.81 0.73 0.73 0.61 0.91 0.75 0.95 0.68 0.71

Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

9

ARTICLE IN PRESS 10

Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

have validated the MODIS aerosol product (Collection 5) using data from the China Aerosol Remote Sensing NETwork (CARSNET) sunphotometers. They found a significant underestimation by the DB algorithm. CONCLUSION We have compared the MODIS derived AOD values (Level 2 version 5.1) from DB algorithm and the conventional DT approach with ground based sunphotometer observations. The study used the AOD data over Ahmedabad, a semiarid location in western India, for the period 2002 to 2005. An important feature noticed is the ability of DB algorithm to retrieve AOD values over the highly reflecting Rann of Kuchchh. This is an understudied area in terms of aerosol climatology, and DB algorithm provides means of investigation of aerosol properties over this region. The correlation of DT - Microtops comparison shows a significant improvement over the previously reported correlation using MODIS Collection 5 data over our study region. The overall correlation is better for DT - Microtops comparison (R2550 = 0.55 for Terra, R2550 = 0.69 for Aqua) than DB Microtops comparison (R2550 = 0.43 for Terra, R2550 = 0.50 for Aqua). Overall, the slopes of MODIS-Microtops comparisons are lower for Terra than Aqua, with the DT slope values being even lower than DB slope values. It implies that there is a diurnal dependence of the error due to improper aerosol model assumption, and that the error is larger in the case of DT derived AOD. AOD values derived by DT approach and DB algorithm are observed to be consistent in case of Aqua MODIS. Except monsoon, the comparison between DB derived AOD and DT derived AOD is superior for Aqua MODIS than Terra MODIS. No dependence of the MODIS -Microtops AOD correlation on surface reflectance was observed. In summary, overall performance of DT approach is better than DB algorithm, except for the Rann of Kuchchh where DT approach faces difficulty in aerosol retrieval. Our recommendation is to use the DB derived AOD over the Rann of Kuchchh region, where DT approach has difficulty retrieving AOD due to higher surface reflectance. For other regions, DT derived AOD should be used. Thus, the best approach would be to create a merged data set from the two approaches and use it to study the aerosol climatology over the Gujarat region. The results obtained in this work would aid in improvement of the MODIS aerosol product with updated version of the algorithms. ACKNOWLEDGEMENTS We acknowledge the Land Processes Distributed Active Archive Centre (LPDAAC) for providing the MODIS aerosol product used in this work. We thank Michael D. King and N. Christina Hsu, NASA-GSFC, for providing updates and information on the MODIS Deep Blue aerosol product. REFERENCES Choudhry, P., Misra, A. and Tripathi, S.N. (2012). Study

of MODIS Derived AOD at Three Different Locations in the Indo Gangetic Plain: Kanpur, Gandhi College and Nainital. Ann. Geophys. 30: 1479–1493, doi: 10.5194/an geo-30-1479-2012. Chu, D.A., Kaufman, Y.J., Ichoku, C., Remer, L.A., Tanre, D. and Holben, B.N. (2002). Validation of MODIS Aerosol Optical Depth Retrieval over Land. Geophys. Res. Lett. 29: 8007, doi: 10.1029/2001GL013205. Dave, J.V. (1970). Intensity and Polarization of the Radiation Emerging from a Plane-parallel Atmosphere Containing Monodispersed Aerosols. Appl. Opt. 9: 2673–2684, doi: 10.1364/AO.9.002673. Dubovik, O., Holben, B., Eck, T.F., Smirnov, A., Kaufman, Y.J., King, M.D., Tanre, D. and Slutsker, I. (2002). Variability of Absorption and Optical Properties of Key Aerosol Types Observed in Worldwide Locations. J. Atmos. Sci. 59: 590–608, doi: 10.1175/1520-0469(2002) 0592.0.CO;2. Ganguly, D., Jayaraman, A. and Gadhavi, H. (2006). Physical and Optical Properties of Aerosols over an Urban Location in Western India: Seasonal Variabilities. J. Geophys. Res. 111: 24206, doi: 10.1029/2006JD007392. Gatebe, C.K., King, M.D., Tsay, S.C., Ji, Q., Arnold, G.T. and Li, J.Y. (2001). Sensitivity of Off-nadir Zenith Angles to Correlation between Visible and Near-infrared Reflectance for Use in Remote Sensing of Aerosol over Land. IEEE Trans. Geosci. Remote Sens. 39: 805–819, doi: 10.1109/36.917901. Gautam, R., Hsu, N.C. and Lau, K.M. (2010). Premonsoon Aerosol Characterization and Radiative Effects over the Indo-Gangetic Plains: Implications for Regional Climate Warming J. Geophys. Res. 115, 17208, doi: 10.1029/ 2010JD013819. Gautam, R., Hsu, N.C., Tsay, S.C., Lau, K.M., Holben, B., Bell, S., Smirnov, A., Li, C., Hansell, R., Ji, Q., Payra, S., Aryal, D., Kayastha, R. and Kim, K.M. (2011). Accumulation of Aerosols over the Indo-Gangetic Plains and Southern Slopes of the Himalayas: Distribution, Properties and Radiative Effects during the 2009 Premonsoon Season. Atmos. Chem. Phys. 11: 12841–12863, doi: 10.5194/acp-11-12841-2011. Hatzianastassiou, N., Matsoukas, C., Drakakis, E., Stackhouse, Jr., P.W., Koepke, P., Fotiadi, A., Pavlakis, K.G. and Vardavas, I. (2007). The Direct Effect of Aerosols on Solar Radiation Based on Satellite Observations, Reanalysis Datasets, and Spectral Aerosol Optical Properties from Global Aerosol Data Set (GADS). Atmos. Chem. Phys. 7: 2585–2599. Haywood, J.M. and Ramaswamy, V. (1998). Global Sensitivity Studies of the Direct Radiative Forcing due to Anthropogenic Sulfate and Black Carbon Aerosols. J. Geophys. Res. 103, 6043–6058, doi: 10.1029/97JD03426. Hsu, N.C., Tsay, S.C., King, M.D. and Herman, J.R. (2004). Aerosol Properties over Bright-Reecting Source Regions. IEEE Trans. Geosci. Remote Sens. 42: 557–569, doi: 10.1109/TGRS.2004.824067. Ichoku, C., Chu, D.A., Mattoo, S., Kaufman, Y.J., Remer, L.A., Tanre, D., Slutsker, I. and Holben, B.N. (2002). A Spatio-temporal Approach for Global Validation and

ARTICLE IN PRESS Misra et al., Aerosol and Air Quality Research, x: 1–11, xxxx 

Analysis of MODIS Aerosol Products. Geophys. Res. Lett. 29: 8006, doi: 10.1029/2001GL013206. Jethva, H., Satheesh, S.K. and Srinivasan, J. (2007a). Assessment of Second-generation MODIS Aerosol Retrieval (Collection 005) at Kanpur, India. Geophys. Res. Lett. 34: 19802, doi: 10.1029/2007GL029647. Jethva, H., Satheesh, S.K. and Srinivasan, J. (2007b). Evaluation of Moderate Resolution Imaging Spectroradiometer (MODIS) Collection 004 (C004) Aerosol Retrievals at Kanpur, Indo-Gangetic Basin. J. Geophys. Res. 112: 14216, doi: 10.1029/2006JD007929. Kaufman, Y.J., Tanre, D., Remer, L.A., Vermote, E.F., Chu, A. and Holben, B.N. (1997a). Operational Remote Sensing of Tropospheric Aerosol over Land from EOS Moderate Resolution Imaging Spectroradiometer. J. Geophys. Res. 102, 17051, doi: 10.1029/96JD03988. Kaufman, Y.J., Wald, A.E., Remer, L.A., Gao, B.C., Li, R.R. and Flynn, L. (1997b). The MODIS 2.1-μm Channelcorrelation with Visible Reflectance for Use in Remote Sensing of Aerosol. IEEE Trans. Geosci. Remote Sens. 35: 1286–1298. King, M.D., Kaufman, Y.J., Tanre, D. and Nakajima, T., (1999). Remote Sensing of Tropospheric Aerosols from Space: Past, Present, and Future. Bull. Am. Meteorol. Soc. 80, 2229–2260, doi: 10.1175/1520-0477(1999)080 2.0.CO;2. Levy, R.C., Remer, L.A., Martins, J.V., Kaufman, Y.J., Plana-Fattori, A., Redemann, J. and Wenny, B. (2005). Evaluation of the MODIS Aerosol Retrievals over Ocean and Land during CLAMS. J. Atmos. Sci. 62: 974– 992, doi: 10.1175/JAS3391.1. Levy, R.C., Remer, L.A. and Dubovik, O. (2007a). Global Aerosol Optical Properties and Application to Moderate Resolution Imaging Spectroradiometer Aerosol Retrieval over Land. J. Geophys. Res. 112: 13210, doi: 10.1029/ 2006JD007815. Levy, R.C., Remer, L.A., Mattoo, S., Vermote, E.F. and Kaufman, Y.J. (2007b). Second-generation Operational Algorithm: Retrieval of Aerosol Properties over Land from Inversion of Moderate Resolution Imaging Spectroradiometer Spectral Reflectance. J. Geophys. Res. 112: 13211, doi: 10.1029/2006JD007811. Levy, R.C., Remer, L.A., Kleidman, R.G., Mattoo, S., Ichoku, C., Kahn, R. and Eck, T.F. (2010). Global Evaluation of the Collection 5 MODIS Dark-target Aerosol Products over Land. Atmos. Chem. Phys. 10: 10399–10420, doi: 10.5194/acp-10-10399-2010. Li, X., Xia, X., Wang, S., Mao, J. and Liu, Y. (2012). Validation of MODIS and Deep Blue Aerosol Optical

11

Depth Retrievals in an Arid/Semi-arid Region of Northwest China. Particuology 10: 132–139. Misra, A., Jayaraman, A. and Ganguly, D. (2008). Validation of MODIS Derived Aerosol Optical Depth over Western India. J. Geophys. Res. 113: 4203, doi: 10.1029/2007JD009075. Morys, M., Mims, F.M., Hagerup, S., Anderson, S.E., Baker, A., Kia, J. and Walkup, T. (2001). Design, Calibration, and Performance of MICROTOPS II Handheld Ozone Monitor and Sun Photometer. J. Geophys. Res. 106: 14573–14582. Prasad, A.K. and Singh, R.P. (2007). Comparison of MISRMODIS Aerosol Optical Depth over the Indo-Gangetic Basin during the Winter and Summer Seasons (20002005). Remote Sens. Environ. 107: 109–119. Remer, L.A., Wald, A.E. and Kaufman, Y.J. (2001). Angular and Seasonal Variation of Spectral Surface Reectance Ratios: Implications for the Remote Sensing of Aerosol over Land. IEEE Trans. Geosci. Remote Sens. 39: 275–283, doi: 10.1109/36.905235. Remer, L.A., Kaufman, Y.J., Tanré, D., Mattoo, S., Chu, D.A., Martins, J.V., Li, R.R., Ichoku, C., Levy, R.C., Kleidman, R.G., Eck, T.F., Vermote, E. and Holben, B.N. (2005). The MODIS Aerosol Algorithm, Products, and Validation. J. Atmos. Sci. 62: 947–973, doi: 10.1175/JAS3385.1. Shi, Y., Zhang, J., Reid, J.S., Hyer, E.J. and Hsu, N.C. (2013). Critical Evaluation of the MODIS Deep Blue Aerosol Optical Depth Product for Data Assimilation over North Africa. Atmos. Meas. Tech. 6: 949–969, doi: 10.5194/amt-6- 949-2013. Tanre, D., Kaufman, Y.J., Herman, M. and Mattoo, S. (1997). Remote Sensing of Aerosol Properties over Oceans Using the MODIS/EOS Spectral Radiances. J. Geophys. Res. 102: 16971, doi: 10.1029/96JD03437. Tripathi, S.N., Dey, S., Chandel, A., Srivastava, S., Singh, R.P. and Holben, B.N. (2005). Comparison of MODIS and AERONET Derived Aerosol Optical Depth over the Ganga Basin, India. Ann. Geophys. 23: 1093–1101, doi: 10.5194/angeo-23-1093-2005. Xie, Y., Zhang, Y., Xiong, X., Qu, J.J. and Che, H. (2011). Validation of MODIS Aerosol Optical Depth Product over China Using CARSNET Measurements. Atmos. Environ. 45: 5970–5978, doi: 10.1016/j.atmosenv.2011. 08.002. Received for review, January 5, 2014 Accepted, April 2, 2014