Aerosol parameters retrieved by Aerosol Robotic Network (AERONET) Sun photometer measurements at the Physics Department of Lecce's University (40°20′N, 18°6′E) are compared to similar Moderate Resolution Imaging Spectroradiometer (MODIS) data retrieved at different spatial resolutions colocated in space and time to contribute to the validation of MODIS aerosol products over southeast Italy and to investigate the correlation dependence on spatial resolution and identify regional biases of Lecce's AERONET data. In particular, MODIS aerosol optical depths retrieved at 550 nm over ocean and over land-ocean for window sizes of 50 × 50, 100 × 100, and 300 × 300 km2 centered on the AERONET monitoring site are correlated to AERONET aerosol optical depths colocated in time. It is shown that correlation factors of linear regressions span the 0.88–0.83 range and weakly tend to reduce with window size. In addition, MODIS aerosol optical depths meet expected uncertainties, and the percentage of data points within expected uncertainties is not affected by the window size. Slope and intercept values of linear regressions fitting over ocean aerosol optical depths are instead different than those fitting over land-ocean aerosol optical depths and are dependent on window size. The observed dependence is analyzed and discussed in the paper. Finally, it is shown that the paper's results can allow inferring that the tested AERONET aerosol parameters can be considered representative, at least, of a ∼300 × 300 km2 southeast Italy area centered on the Lecce's AERONET site.
 Atmospheric aerosols are one of the most important parameters affecting the earth's energy balance and hydrological cycle [Remer et al., 2005]. Depending on their chemical and physical properties, they scatter and/or absorb earthbound solar radiation as well as emitted and reflected radiation from the earth. Aerosols, besides reducing visibility, also interact with cloud droplets, modifying their microphysical properties, thereby influencing their radiative properties and hence precipitation processes. The understanding of aerosols effects on climate is particularly difficult because aerosols are of quite different type and shape, ranging from desert dust to urban pollution, and because aerosol concentrations vary strongly over time and space. Aerosol parameters can be measured in situ or by remote sensing from ground, aircraft, or satellite. All these methods are important and complementary. The application of automatic, ground-based remote-sensing techniques to investigate aerosol effects on climate has advanced significantly in the last years. The Aerosol Robotic Network (AERONET), which is an international network coordinated by NASA Goddard Space Flight Center that maintains more than 200 automatic Sun/sky radiometers worldwide, represents one of the scientific community's efforts to reduce existing uncertainties in aerosol-forcing estimates [Holben et al., 1998]. In contrast to ground-based instruments that acquire continuous measurements at fixed locations, polar-orbiting Sun-synchronous satellite sensors provide a global coverage at nearly constant local solar times, once or twice a day in tropics to midlatitude and multiple overpasses in polar regions [Chu et al., 2003]. Therefore the aerosol remote sensing from long-term operation satellites provides a means to achieve a global and seasonal characterization of aerosols. Satellite sensors provide global images of the entire Earth and allow resolving the spatial patterns resulting from the spatial inhomogeneities of aerosol sources and the temporal patterns resulting from the short lifetimes of aerosols, which are on the order of a few days to a week [Remer et al., 2005]. The last generation of satellites carrying instruments such as the Moderate Resolution Imaging Spectroradiometer (MODIS), the Multiangle Imaging Spectroradiometer (MISR), and the wide field of view imaging radiometer Polarization and Directionally of the Earth's Reflectances reveal (POLDER) the big interest of the scientific community in getting worldwide aerosol characterizations. The Moderate Resolution Imaging Spectroradiometer (MODIS), unlike previous satellite sensors, has the unique ability to retrieve aerosol optical thickness with greater accuracy and to retrieve parameters characterizing aerosol sizes. MODIS aerosol products have recently been validated with ground-based Sun photometer data, particularly those of AERONET [e.g., Chu et al., 2002, 2003; Ichoku et al., 2004; Levy et al., 2005; Remer et al., 2005].
 One of the main objectives of this paper is to investigate the correlation of MODIS aerosol products and corresponding AERONET data retrieved by Sun photometer measurements performed at the Physics Department of Lecce's University (40°20′N, 18°6′E) to contribute to the validation of MODIS retrievals over southeast Italy and hence over the eastern Mediterranean basin. Aerosols from different sources converge to this area: urban/industrial aerosols and seasonal biomass burning from central and eastern Europe, maritime and long-range transported polluted air masses from the Atlantic Ocean, mineral dust from North Africa, and sea spray from the Mediterranean sea itself [De Tomasi et al., 2006]. As a consequence, this area can be well suited to test the performance of MODIS retrieval algorithms. In addition, MODIS aerosol products retrieved for window sizes of 50 × 50, 100 × 100, and 300 × 300 km2 centered on the AERONET monitoring site (Figure 1) are correlated with corresponding AERONET retrievals colocated in time to investigate the dependence of correlation on spatial resolution and hence to understand to what extent the Lecce's AERONET site can be considered representative of a larger area and locally derived aerosol parameters can be of use in General Circulation and Chemical Transport Models [e.g., Kinne et al., 2003; Guibert et al., 2005]. MODIS and AERONET aerosol retrievals from March 2003 to September 2004 are used for the correlation studies reported in this paper. A brief overview of MODIS retrieval algorithms and aerosol products is given in section 2. Few details on AERONET aerosol data are given in section 3. Results are presented in section 4. Summary and conclusion are reported in section 5.
 The Moderate Resolution Imaging Spectroradiometer was launched onboard the EOS Terra and Aqua polar-orbiting satellites since December 1999 and May 2002, respectively [e.g., King et al., 1992]. Terra's orbit around the Earth is timed so that it passes from north to south across the equator in the morning (approximately on 10:30 am), while Aqua passes south to north over the equator in the afternoon (approximately on 1:30 pm). Reflected solar and emitted terrestrial radiances are observed by MODIS in 36 bands ranging from 0.4- to 14.4-μm wavelengths, and two entirely independent algorithms are used to retrieve from MODIS reflectance measurements aerosol parameters over land [Kaufman et al., 1997] and ocean [Tanré et al., 1997] surfaces. Both the land and ocean aerosol algorithms rely on calibrated, geolocated reflectance from the first seven MODIS bands (between 0.47 and 2.1 μm) provided by the MODIS Characterization Support Team (MCST). These reflectance data are first corrected for trace gas and water vapor columns, and final aerosol properties are retrieved for 10 × 10-km boxes. Using results by geolocation and cloud mask data, individual pixels of 10 × 10-km boxes are classified as ocean or land. If all pixels in the 10 × 10-km box are determined to be “ocean”, then the ocean algorithm is performed. If any pixel is land, then the land retrieval is applied. After the land or ocean branch is concluded, the algorithm merges some parameters into combined land and ocean products for convenience [Levy et al., 2005]. The main differences of aerosol retrieval algorithms for land and ocean are as follows: (1) The land algorithm deduces surface reflectivity from the radiance measured at 2.13 μm, whereas the ocean algorithm assumes knowledge of surface reflectivity; (2) the land algorithm fits radiances at two wavelengths, whereas the ocean algorithm fits radiances at six wavelengths to retrieve aerosol products; and (3) the land algorithm distinguishes between dust and nondust, but otherwise uses prescribed combinations of coarse- and fine-mode aerosol for each region whereas, the ocean algorithm solves for both the type and relative amount of coarse- and fine-mode aerosol [Anderson et al., 2005]. The MODIS ocean algorithm, besides aerosol optical depths τ, provides the τ fraction contributed by the fine-mode aerosol and the effective radius of the aerosol distribution [Tanré et al., 1997]. In agreement with theoretical error analysis, τ is derived with an error of Δτ = ±0.03 ± 0.05τ over ocean [Tanré et al., 1997] and with an error of Δτ = ±0.05 ± 0.15τ over land [Chu et al., 1998; King et al., 1999]. Therefore the land and ocean retrieval products are largely independent of each other, and in general, the ocean products are considered much more accurate.
 The measurements reported in this paper are made with the CIMEL Sun/sky radiometer operating at Lecce's University within AERONET since March 2003. The radiometer measures direct Sun radiance at eight spectral channels between 340 and 1020 nm and diffuse sky radiances in the solar almucantar at four wavelengths (441, 673, 873, 1022 nm). The aerosol optical depth processing methodology is described by Holben et al. . In addition, a flexible inversion algorithm, developed by Dubovik and King , is used to retrieve aerosol optical depths, columnar aerosol volume size distributions, refractive indices, and single-scattering albedos from direct Sun and diffuse sky radiance measurements. A brief discussion on the accuracy of the individual retrievals is reported in the work of Dubovik et al. [2000, 2002]. The aerosol optical depth accuracy is of ±0.01 or slightly higher for the field instruments in the range 440–1020 nm and about 0.015–0.02 in the UV [Holben et al., 1998; Eck et al., 1999]. In this paper we use level 1.5 AERONET data which are cloud screened [Smirnov et al., 2000] and, on average, provide aerosol optical depths at 15-min intervals. Some results on the aerosol load characterization by the retrievals of the AERONET Sun/sky radiometer operating at Lecce's University are reported in the works of Perrone et al.  and Tafuro et al. .
4. Methodology and Results
 In this paper the authors use level 2 aerosol products from MODIS onboard the EOS Terra satellite that are operationally derived at 10-km spatial resolution. In order to take into account both spatial and temporal variability of the aerosol distribution, the MODIS retrievals at 10-km spatial resolution and the AERONET direct Sun measurements at 15-min intervals [Holben et al., 1998] need to be colocated in space and time [Chu et al., 2002]. The methodology adopted in this work to compare temporal statistics from AERONET to spatial statistics from MODIS is the same to the one described by Ichoku et al. . We require that at least two out of the possible five AERONET measurements are within ±30 min of MODIS overpasses and that at least five out of the possible 25 MODIS retrievals are in a square box of 50 × 50 km centered over the AERONET site [Chu et al., 2003]. Accordingly, Sun photometer retrievals between ±60 and ±90 min of the satellite overpass are compared with MODIS products in a square box of 100 × 100 km and of 300 × 300 km, respectively, centered on the Sun photometer site. The mean values of the colocated spatial and temporal ensemble are then used to investigate the correlation between AERONET and MODIS aerosol products. Aerosol optical depths and fine fraction parameters provided by MODIS at 550 nm are used for the correlation study of this paper. MODIS and AERONET wavelengths do not match exactly except at 870 nm. According to Remer et al. , the interpolation of AERONET data is done on a log-log plot assuming linearity between 0.44 and 0.87 μm. The error in the interpolation varies between 0 and ∼10% depending on the aerosol type (due to nonlinear spectral dependence), with fine-mode-dominated aerosol at high optical thickness introducing the most error and a mixed- or coarse-dominated aerosol introducing the least [Eck et al., 1999; Ichoku et al., 2005]. As we have mentioned in section 1, MODIS and AERONET aerosol retrievals from March 2003 to September 2004 are used for the correlation studies reported in this paper.
4.1. AERONET Versus MODIS Ocean AODs
 The data analysis that follows uses aerosol products retrieved over oceans from MOD04_L2 scientific data set Effective_Optical_Depth_Average_Ocean files, which can be found at the Web site http//:modis-atmos.gsfc.nasa.gov/. Figures 2a–2c show the scatterplot (open dots) of the mean values of MODIS ocean aerosol optical depths at 550 nm, τM, retrieved for window sizes of (1) 50 × 50, (2) 100 × 100, and (3) 300 × 300 km2, respectively, centered on the AERONET site versus the corresponding mean values of Sun photometer aerosol optical depths, τA, colocated in time. The total number of data points N of each plot of Figures 2a–2c is also given. MODIS standard deviations σM, referring to the data points of Figures 2a–2c versus the corresponding Sun photometer standard deviations σA, are plotted on Figures 2d–2f for window sizes of 50 × 50, 100 × 100, and 300 × 300 km2, respectively.
σM values are representative of both the retrieval uncertainties and the spatial variability of aerosol optical depths; aerosol spatial distributions may vary significantly even within a few kilometers. Besides retrieval uncertainties, σA values are also representative of the temporal variability of the atmospheric aerosol distribution. We observe from Figures 2d–2f that both σM and σA increase with window size. It is possible that the dependence of σM and σA on window size is mainly determined by the higher spatial and temporal variability of the aerosol distribution as averaging area and time are increased. σM and σA reach values up to ∼0.05 when the window size is 50 × 50 km and up to ∼0.2 when the window size gets 300 × 300 km. In addition, Figures 2d–2f indicate that σM are on average larger than σA. Latter results can be due to either the larger aerosol variability with space than with time or to the larger uncertainties of τM (Δτ = ±0.03 ± 0.05τ) with respect to the uncertainties of τA values (±0.01).
 The solid line of each plot of Figures 2a–2c represents the regression line fitting the data points. Regression line parameters and correlation coefficients (R) are reported on the top of each plot. The authors observe that correlation coefficients vary from 0.86 to 0.83. The regression line slope decreases from 0.85 down to 0.7 as the window size area increases from 50 × 50 km2 up to 300 × 300 km2. While the regression line intercept increases from 0.03 up to 0.05 as the window size area is varied from 50 × 50 km2 up to 300 × 300 km2.
 The dashed lines of each plot of Figures 2a–2c represent the MODIS “expected errors” to contain the mean and the first standard deviation (66%) of all points (i.e., Δτ = ±0.03 ± 0.05τ) when the ocean algorithm is used. The constant term represents the estimated error due to the surface reflectance assumption, while the second term that is proportional to τ represents the error due to aerosol model assumptions. The data analysis of Figures 2a–2c shows that 70, 67, and 70% of the τ values, respectively, meet the prespecified accuracy conditions. It is worth noting that the percentage of τ values within prespecified accuracy conditions does not appear significantly affected by the window size. Figure 2a also reveals that the regression line obtained by fitting the data points of the 50 × 50 km2 window size fits inside the expected error lines at least up to τA values of about 0.6. The regression line parameters of each plot of Figures 2a–2c indicate either that MODIS overestimates aerosol optical depths at low aerosol loadings and that MODIS underestimates aerosol optical depths at higher aerosol loadings. In addition, the aerosol optical depth underestimation gets more significant as the window size is increased.
 Aerosol optical depths retrieved at low aerosol loadings are expected to be more affected by the assumed surface reflectivity, while τ values retrieved at high aerosol loadings are expected to be more dependent on the aerosol model. Besides the spatial variability of the aerosol characteristics, the surface reflectivity underestimation, whose effects get more significant with the increase of the averaging area and hence of N, can also be responsible of the intercept positive value of the regression line, which increases with the window size, while the MODIS aerosol model assumption can be responsible of the regression line slope values that are always smaller than unity and take smaller values as the window size is increased. According to Zhao et al. , a slope that is different from unity indicates that there may be some inconsistency between aerosol microphysical and optical properties used in the MODIS retrieval algorithm and those in the real situation.
 The Chesapeake Lighthouse and Aircraft Measurements for Satellites (CLAMS) experiment has recently been designed in part to examine and validate MODIS retrievals over coastlines [e.g., Smith et al., 2005; Levy et al., 2005]. The measurements that have been performed during July–August 2001 have revealed that the optical depths over the ocean retrieved from MODIS compared well to those measured by the Sun photometers; 86% of all individual ocean retrievals at 550 nm were within the expected error lines. However, it is worth noting that the scatterplot of MODIS versus Sun photometer aerosol optical depths at 550 nm has been fitted by a regression line with a positive intercept value (equal to 0.02) and a slope different from unity and equal to 0.92 [Levy et al., 2005]. The scatterplot of 2052 MODIS aerosol optical depths at 550 nm colocated with AERONET aerosol optical depths retrieved at different stations either on the coast or on an island has also revealed either that the MODIS ocean algorithm leads to a slight under prediction at high optical depths [Remer et al., 2005] and that 62% of all retrievals over ocean at 550 nm were within the defined expected uncertainty. The quality of the aerosol optical depth data retrieved over ocean by MODIS from 2000 to 2003 has also been evaluated by Ichoku et al.  by investigating the correlation of MODIS and AERONET data colocated in time and referring to several worldwide sites. Despite the results of the works by Remer et al.  and Levy et al. , the Ichoku et al.  study has provided results similar to those of this paper.
Figure 3a shows the τM frequency distribution referring to the 50 × 50 km2 (solid line), 100 × 100 km2 (dashed line), and 300 × 300 km2 (grey solid line) window size. Figure 3b shows the frequency distribution of the mean aerosol optical depths calculated by averaging Sun photometer τ values retrieved between ±30 min (solid line), ±60 min (dashed line), and ±90 min (grey solid line) of the Terra satellite overpass. We observe from Figure 3a that the τM frequency distribution is not significantly affected by window size and that all frequency distributions are peaked at about 0.08. In addition, Figure 3b reveals that the τA frequency distribution is not significantly affected by the averaging time and that the plots of Figure 3b are similar to those of Figure 3a.
 Mean optical depths and and corresponding standard deviations for different spatial and temporal resolutions, respectively, are summarized on Table 1 to facilitate comparison between MODIS and AERONET data.
Table 1. Mean AODs and Corresponding Standard Deviations for Different Spatial Resolutions by MODIS and for Different Temporal Resolutions by AERONETa
Spatial Resolution (km2)
represents mean MODIS ocean AODs and represents corresponding mean AERONET AODs colocated in time. represents mean MODIS land-ocean AODs and represent corresponding mean AERONET AODs colocated in time.
50 × 50
0.2 ± 0.1
0.3 ± 0.1
0.2 ± 0.1
0.2 ± 0.1
100 × 100
0.19 ± 0.09
0.3 ± 0.1
0.2 ± 0.1
0.2 ± 0.1
300 × 300
0.19 ± 0.09
0.3 ± 0.1
0.2 ± 0.1
0.2 ± 0.1
 In order to investigate if the correlation between MODIS and AERONET aerosol optical depths is dependent on the time of the year, we have plotted in Figures 4a–4cτM (black dots) and colocated in time τA (grey dots) values as a function of the time of the year and for the (1) 50 × 50, (2) 100 × 100, and (3) 300 × 300 km2 window sizes, respectively. Open black and open grey dots represent monthly average values of τM and τA, respectively. Figures 4a–4c reveal that τM closely follows the temporal evolution of τA at all tested window sizes. The latter comment is further supported by Figures 5a–5c showing the differences ΔτM−A (grey full dots) between monthly averaged values of τM and τA. The solid black line represents ΔτM−A = 0 while dotted lines represent MODIS “expected errors.” Figures 5a–5c show that most of the ΔτM−A values fit inside the expected error lines at all tested window sizes. Grey full dots above or below the solid black line represent MODIS overestimation or underestimation, respectively, with respect to AERONET values. Latter results indicate that, over southeast Italy, the differences ΔτM−A are not significantly affected by the time of the year at all tested window sizes.
 In conclusion, the results of this section at first show that the correlation of MODIS and AERONET aerosol optical depths is weakly affected by the window size; 70, 67, and 70% of τ values retrieved at 50 × 50, 100 × 100, and 300 × 300 km2 window sizes, respectively, meet prespecified accuracy conditions. In addition, aerosol optical depth frequency distributions (Figure 3a) and temporal evolutions (Figures 4a–4c) do not appear significantly affected by the window size. As a consequence, the authors believe that the AERONET aerosol optical depths retrieved at Lecce can be considered representative of an area of at least 300 × 300 km2; the Sun photometer continues to describe the evolution of regional aerosol characteristics, but it begins to lose predictive capabilities since slopes and intercepts get worse as the window size is increased. It is possible that the dependence on window size of intercept and slope values (Figures 2a–2c) is weakly determined by the larger spatial and temporal variability of aerosol properties with window size and that it is mostly due to the MODIS surface reflectance underestimation and to some inconsistency of the MODIS ocean algorithm; last two effects are also expected to increase with window size and hence with the number of data point increase. The authors believe that, besides this paper's results, the data reported by Levy et al. , Remer et al. , and Ichoku et al.  can demonstrate that the ground surface reflectance and the aerosol properties of the sites chosen for the correlation study affect intercept and slope values of the regression lines fitting the scatterplots of Sun photometer and MODIS aerosol optical depths.
4.2. AERONET and MODIS Land-Ocean AODs
 As it has been told, two entirely independent algorithms, the ocean and land algorithm, are used for deriving aerosol products over land and over ocean, respectively. However, after the land or ocean branch is concluded, the algorithm merges some parameters into combined land and ocean products for convenience [Levy et al., 2005]. The main combined land-ocean products are the aerosol optical depth and the fine fraction parameter at 550 nm. Note that the land and ocean retrievals are not required to meet at the shoreline and may be discontinuous. Considering the peculiar location of the Lecce's University AERONET site that is on a narrow peninsula of southeast Italy (Figure 1), we consider rather meaningful to investigate in this section the correlation of AERONET and MODIS land-ocean aerosol optical depths (MOD04_L2 Scientific Data Set: Optical_Depth_Land_And_Ocean) colocated in time and space. Figures 6a–6c show the scatterplot (open dots) of the mean values of MODIS land-ocean aerosol optical depths τ*M retrieved for window sizes of 50 × 50, 100 × 100, and 300 × 300 km2 centered on the AERONET site versus the mean values of corresponding Sun photometer aerosol optical depths τ*A colocated in time. Solid lines represent regression lines fitting the data points while dashed lines represent prelaunch expected uncertainties over land that are given by the relationship Δτ = ±0.05 ± 0.15τ [Chu et al., 1998; King et al., 1999]. It is worth noting that there are no quantified expected uncertainties for land-ocean aerosol optical depths. Scatterplots of standard deviations of land-ocean data pointsσ*M versus corresponding σ*A values are plotted on Figures 6d–6f.
 The comparison of Figures 2a–2c and Figures 6a–6c at first reveals that the data point dispersion is larger on Figures 6a–6c. In addition, Figures 6d–6f reveal a small trend of σ*M to increase with window size and that the σ*M variability range is nearly twice larger than that of σ*A:σ*M values up about 0.25 are retrieved at 50 × 50 and 300 × 300 km2 window sizes. Besides the larger uncertainties of aerosol optical depths retrieved over land, the larger surface heterogeneity can be responsible of latter results; the land-ocean retrievals are not required to meet at the shoreline and may be discontinuous. Correlation coefficients of the plots of Figures 6a–6c vary from 0.88 to 0.84 as the window size is increased from 50 × 50 to 300 × 300 km2, and the percentages of data points within expected uncertainties are 85, 88, and 82% for the 50 × 50, 100 × 100, and 300 × 300 km2 window sizes, respectively. As the window size increases, the proportion of land and ocean contribution varies, and this may also account for latter results. In accordance to the results of Figures 2a–2c, regression line parameters of Figures 6a–6c also indicate that MODIS overestimates and underestimates aerosol optical depths at low and high aerosol loadings, respectively, and that both effects get more significant as the window size is increased. However, regression line slopes values of Figures 6a–c are closer to unity than those of Figures 2a–2c, and these results may indicate that the MODIS land-ocean aerosol optical depths are best suited to represent the aerosol properties over southeast Italy. On the contrary, the higher intercept values of the regression lines of Figures 6a–6c, with respect to those of Figures 2a–2c, likely indicate that the ground reflectivity underestimation is larger for the MODIS land retrieval algorithm over southeast Italy. Intercept values of Figures 6a–6c are also greater than the expected offset of 0.05 for all tested window sizes. Figures 7a–7c, showing τ*M (black dots) and τ*A (grey dots) and the corresponding monthly means (open dots) as a function the time of the year, reveal that τ*M closely follows the temporal evolution of τ*A at all tested window sizes, but τ*M monthly means are larger than τ*A during all the year of almost a constant amount. The latter comment is further supported by Figures 8a–8c showing the differences Δτ*M−A (grey full dots) between monthly averaged values of τ*M and τ*A corresponding values. The solid black line represents Δτ*M−A = 0, while dotted lines represent MODIS “expected errors.” Δτ*M−A values mostly fit inside expected error lines at all tested window sizes, but they are almost above solid black lines since MODIS overestimates aerosol optical depths with respect to AERONET values. The land validation study of Remer et al.  also reveals that the plot of 5906 MODIS aerosol optical depths at 550 nm colocated with AERONET measurements at different sites shows a positive offset of 0.068 that is larger than expected. In addition, the correlation study reported by Ichoku et al.  shows that, over land and at 550 nm, the scatterplot of level 1.5 AERONET data and Terra-MODIS aerosol optical depths (T004) is fitted by a regression line with 0.72 slope value, 0.128 intercept value, and 0.68 linear correlation coefficient and that only 53.5% of 9740 data points were within expected error bounds. The correlation at 550 nm of MODIS versus Sun photometer aerosol optical depths over land during CLAMS [Levy et al., 2005] is disappointing; the regression line intercept and slope are equal to 0.21 and 0.64, respectively, and the correlation coefficient takes the value of 0.36. The land algorithm deduces ground reflectivity from the radiance measured at 2.13 μm. Both measurements [Kaufman and Remer, 1994; Kaufman et al., 1997] and theoretical studies [Kaufman et al., 2002] have demonstrated that, for certain vegetated surfaces throughout the globe, the surface reflectance at 0.47 and 0.66 μm can be derived from the mean radiance at 2.13 μm by empirical relationships. However, it has also been shown that these relationships do not hold for specific regions, and it is possible that they are responsible of the positive bias revealed either by the scatterplots of Figures 6a–6c or by the observation worldwide [Remer et al., 2005; Levy et al., 2005; Ichoku et al. 2005].
 Finally, Figures 9a and 9b that show the τ*M and τ*A frequency distribution, respectively, reveal that the τ*M and τ*A frequency distributions are not significantly affected by the window size. In accordance to the comments of the previous paragraph, the results of this section also show that the AERONET aerosol optical depths retrieved at Lecce can be considered representative of an area of at least 300 × 300 km2.
Table 1 provides mean optical depths and and corresponding standard deviations for different spatial and temporal resolutions, respectively.
4.3. AERONET and MODIS Ocean Fine Fraction Parameters
 According to Tanré et al. , Levy et al. , and Remer et al. , the six reflectances measured from MODIS in the 0.55–2.13 μm spectral range and used in the ocean retrieval, upon finding best fits to modeled reflectances, allow getting the total optical depth at one wavelength and the fine fraction parameter ηM at that wavelength. ηM represents the ratio between fine mode and total optical thickness. The MODIS ocean inversion algorithm is based on Lookup Tables (LUT) that consists of four fine modes and five coarse modes [Remer et al., 2005]. The procedure requires both a fine mode and a coarse mode for each retrieval [Kleidman et al., 2005]. The modes from the LUT are combined using ηM as the weighting parameter. For each of the 20 combinations of one fine mode and one coarse mode, the inversion finds the pair of τM and ηM that gives a residual error ɛ < 3%. A detailed description of the strategy for inversion of the MODIS spectral data is given by Tanré et al. .
Figures 10a–10b show versus the time of the year the available ηM values at 550 nm (grey dots) from March 2003 to September 2004 (MOD04_L2 Scientific Data Set: Optical_depth_ratio_small_ocean_055 micron) for the 50 × 50 km and the 300 × 300 km window size centered on the AERONET site, respectively. For comparison Figure 10c provides ηA versus the time of the year. ηA is the ratio between fine mode and total optical thickness at 550 nm (grey dots) retrieved from March 2003 to September 2004 AERONET measurements. The AERONET inversion algorithm allows retrieving aerosol volume distributions, and all particles with the radius smaller than 0.6 μm are considered fine while those with the radius larger than 0.6 μm are considered coarse. Therefore the AERONET fine-mode optical thickness represents the aerosol optical depth due to particles with the radius smaller than 0.6 μm. Hence ηM and ηA do not represent the same aerosol parameter, but they are comparable parameters, and the authors believe that it is more meaningful to compare the temporal evolution of both parameters instead of plotting simultaneous data in a scatterplot. In addition, Remer et al.  have shown that rather few simultaneous ηM and ηA data can be available if 1 or 2 years of AERONET and MODIS aerosol products are used for comparison. ηA is retrieved from sky radiance measurements that are taken less often than direct Sun measurements in the AERONET protocol [Dubovik et al., 2000]. The monthly distribution of data points n (grey lines) is also shown on each plot of Figure 10. Black full dots and error bars represent in Figure 10 monthly average values of the fine fraction optical parameter and corresponding standard deviations; the latter indicate the fine fraction parameter variability range. The fine fraction contribution to the total optical thickness is an important parameter to assess the climate impact of anthropogenic aerosols [Kaufman et al., 2002]; it helps us to discriminate natural aerosols (largely, mechanically generated dust and sea salt) from anthropogenic ones (largely, combustion-generated sulfates, organics, and black carbon). The comparison of Figure 10a and 10b at first reveals that the ηM temporal evolution is not affected by the window size; monthly average values of the 50 × 50 km window size are rather similar to those of the 300 × 300 km window size. In addition, the authors observe that ηM takes values in the 0.7–0.8 and 0.4–0.6 range from April to September and from October to March, respectively. On the contrary, ηA monthly means (Figure 10c) span the 0.7–0.8 range during the whole year and are not significantly affected by seasons. Perrone et al.  allow understanding the ηA temporal evolution. In Figure 9 of that paper, given is the temporal plot of the ratio between fine and coarse number of particles per cross section of the atmospheric column (Nf/Nc). The plot shows that the Nf/Nc ratio is characterized by a marked seasonal evolution; fine particles dominate during the year and mainly on spring-summer, Nf/Nc monthly means reach values larger than ∼3 × 103 from June to September. Therefore if the aerosol optical depth due to fine-mode particles is significantly larger than that due to coarse-mode particles, ηA that is defined as the ratio between the fine-particle aerosol optical depth and the total (fine + coarse particles) optical depth is expected to be not significantly affected by season.
 The differences ΔηM−A between ηM and ηA monthly means for the 50 × 50 and 300 × 300 km window size are shown on Figure 11a and 11b, respectively. Dashed lines are the prelaunch estimated uncertainties, ±30% [Remer et al., 2005]. Figure 11 reveals that spring-summer ∣ΔηM−A∣ values are within prelaunch estimated uncertainties and are lower than 14% from April to September. On the contrary, ∣ΔηM−A∣ takes values larger than the prelaunch estimated uncertainties on winter months. These differences may either be physically real and caused by problems in comparing temporal data with spatial data or may merely be an artifact of problems associated with the MODIS retrieval algorithm as it is explained below.
 The data point distribution of Figures 10a and 10b show that ηM monthly values of autumn-winter months are determined by averaging rather few data points, mainly in November and December. In addition, aerosol optical depths are smaller on autumn-winter months (Figure 4), and according to Remer et al. , at low aerosol optical depths, because of less signal, there is greater susceptibility to all algorithmic and sensor uncertainties. In accordance to the above discussion, it is possible to assume that the autumn-winter ηM values are affected by larger uncertainties and in particular that the MODIS ocean algorithm underestimates the fine fraction contribution on autumn-winter. However, the autumn-winter regional variation of the aerosol properties may also be responsible of latter results. It could be possible that, especially on winter, when the wide spread of regional haze is probably not developed, the aerosol load over the sea seen by MODIS is more affected by oceanic aerosols than the aerosol load on land monitored by AERONET, being the latter probably more affected by local land sources [De Tomasi et al., 2006].
Remer et al.  have recently reported comparisons of monthly means of MODIS- and AERONET-derived fine fraction values for different worldwide sites. They show that for some sites, such as GSFC, Anmyon, and Male, ηA and ηM agree to within 20% for much of the year. For Bermuda, Midway Island, and Lanai, the agreement is sustained for the first 6 months of the year until ηM drops down to much lower values. Latter results have been ascribed by Remer et al.  to the larger uncertainties of aerosol size parameters retrieved at low aerosol loads. However, it is worth noting from the data reported by Remer et al.  that ∣ΔηM−A∣ takes values much larger than the expected accuracy at the sites where the temporal evolution of ηA is not significantly affected by seasons, like Bermuda, Midway Island, Lanai, and Rome-Tor-Vergata.
5. Summary and Conclusion
 Aerosol parameters retrieved by AERONET Sun photometer measurements at the Physics Department of Lecce's University from March 2003 to September 2004 are compared in this paper to corresponding MODIS data retrieved at different spatial resolutions colocated in space and time, to contribute to the validation of MODIS aerosol products over southeast Italy, to investigate the correlation dependence on spatial resolution, and to identify regional biases of Lecce's AERONET data. Despite most of the studies on the validation of MODIS retrievals, the results of this paper refer to a single site on southeast Italy where different aerosol types may converge during the year and many aerosol types can superimpose mainly in summer as a consequence of weather stability [Perrone et al., 2005]. Then, the area can be well suited to test the performance of MODIS retrieval algorithms. Averaged values of ocean and land-ocean MODIS aerosol optical depths retrieved at 550 nm for window sizes of 50 × 50, 100 × 100, and 300 × 300 km2 centered on Lecce have been correlated to AERONET aerosol optical depths colocated in time. The authors have observed that correlation factors of linear regressions span the 0.88–0.83 range and weakly tend to reduce with the window size increase. In addition, MODIS aerosol optical depths meet expected uncertainties; 70, 67, and 70% of τM values of the 50 × 50, 100 × 100, and 300 × 300 km2 window size, respectively, is within expected uncertainties, while 85, 88, and 82% of τ*M values retrieved at 50 × 50, 100 × 100, and 300 × 300 km2 window size, respectively, meets prespecified accuracy conditions. In addition, the authors have observed that frequency distributions and temporal evolutions of ocean and land-ocean mean aerosol optical depths are not dependent on window size. All these results can allow inferring that AERONET aerosol optical depths retrieved at Lecce can be considered representative at least of a 300 × 300 km2 area centered on Lecce and hence that locally derived aerosol parameters can be of use in General Circulation and Chemical Transport Models based on spatial resolutions of few hundred kilometers. This represents one of the main issues of the paper, and it can be mainly due to the geographical location of Lecce's AERONET site that is on a flat area away from large sources of local pollution.
 The authors have also found that slopes and intercepts of the linear regressions fitting the aerosol optical depth scatterplots are dependent on MODIS retrieval algorithm and window size. But regression lines fitting ocean and land-ocean MODIS optical depths values are all characterized by a positive intercept value, which weakly increases with window size, and by a slope value smaller than unity that decreases as the window size increases. As a consequence, the authors at first have observed that MODIS overestimates τ at low aerosol loadings, and it is possible that this result is due to the fact that both the MODIS ocean and mainly the MODIS land algorithm underestimate the ground surface reflectance. Similar results have been reported in several papers. The authors have also found that the slopes of the regression lines fitting the scatterplots with ocean and land-ocean MODIS aerosol optical depths vary within the 0.85–0.70 and 0.95–0.85 range, respectively. As it has been told MODIS aerosol model assumptions are considered responsible of regression line slope values different from unity. Then, being the slopes of the regression lines fitting the scatterplots with land-ocean MODIS aerosol optical depths closer to unity, the paper's results can indicate that the land-ocean MODIS aerosol optical depths better represent the aerosol properties over southeast Italy.
 Finally, the authors have observed that the monthly evolution of MODIS and AERONET aerosol optical depths is rather similar, and it is characterized by a significant seasonal dependence. The temporal evolution of the MODIS fine fraction parameter ηM and of AERONET ηA values has instead revealed that ηM monthly means depend on seasons and take values in the 0.7–0.8 and 0.4–0.6 range in spring-summer and autumn-winter, respectively. On the contrary, ηA monthly means span the 0.7–0.8 range during all year. It has been shown that it is possible that the marked seasonal evolution of ηM is due either to the autumn-winter regional variation of the aerosol properties or to the MODIS ocean algorithm that underestimates the fine fraction contribution on autumn-winter months because of the lower aerosol loads and hence the less signal and greater susceptibility to all algorithmic and sensor uncertainties.
 In conclusion, the authors believe that the paper's results can contribute to the validation of MODIS aerosol retrievals over the coastal sites of southeast Italy and hence of the eastern Mediterranean sea. This paper for the first time provides some results on the correlation of MODIS and AERONET aerosol products over southeast Italy.
 The authors would like to thank MODIS science data support team for processing MODIS data. Work supported by Ministero dell'Istruzione dell'Università e della Ricerca (contract 2004023854).