An evaluation of satellite aerosol products against sunphotometer measurements

Remote Sensing of Environment 115 (2011) 3102–3111

Contents lists available at ScienceDirect

Remote Sensing of Environment j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e

An evaluation of satellite aerosol products against sunphotometer measurements Francois-Marie Bréon a,⁎, Anne Vermeulen b, Jacques Descloitres b a b

Laboratoire des Sciences du Climat et de l'Environnement, Laboratoire CEA-CNRS-UVSQ, CEA/DSM/LSCE, 91191 Gif sur Yvette, France ICARE Data and Services Center, Université Lille 1, 59650 Villeneuve d'Ascq, France

a r t i c l e

i n f o

Article history: Received 6 October 2010 Received in revised form 20 June 2011 Accepted 22 June 2011 Available online 23 July 2011 Keywords: Aerosol Remote sensing Validation Atmosphere Sunphotometer

a b s t r a c t Because atmospheric aerosols scatter sunlight back to space, reflectance measurements from spaceborne radiometers can be used to estimate the aerosol load and its optical properties. Several aerosol products have been generated in a systematic way, and are available for further studies. In this paper, we evaluate the accuracy of such aerosol products derived from the measurements of POLDER, MODIS, MERIS, SEVIRI and CALIOP, through a statistical comparison with Aerosol Optical Depth (AOD) measurements from the AERONET sunphotometer network. Although this method is commonly used, this study is, to our knowledge, among the most extensive of its type since it compares the performance of the products from 5 different sensors using up to five years of data for each of them at global scale. The choice of these satellite aerosol datasets was based on their availability at the ICARE Data and Service Centre (www.icare.univ-lille1.fr). We distinguish between retrievals over land and ocean and between estimates of total and fine mode AOD. Over the oceans, POLDER and MODIS retrievals are of similar quality, with RMS difference lower than 0.1 and a correlation with AERONET of around 0.9. The POLDER estimates suffer from a small positive bias for clean atmospheres, which weakens its statistics. The other aerosol products are of lesser quality, although the SEVIRI products may be of interest for some applications that require a high temporal resolution. The MERIS product shows a very high bias. Over land, only the MODIS product offers a reliable estimate of the total AOD. On the other hand, the polarization-based retrieval using POLDER data allows a better fine mode estimate than that from MODIS. These results suggest the need for a product combining POLDER and MODIS products over land. The paper also analyses how the statistics change with the spatial and temporal thresholds that are used. Spatio-temporal averaging improves the statistics only slightly, which indicates that random errors are not dominant in the error budget. The paper includes various statistical indicators at global scale, and detailed results at individual ground stations can be obtained on request from the authors. © 2011 Elsevier Inc. All rights reserved.

1. Introduction Atmospheric aerosol is an essential component of the climate system and is recognized as the primary uncertainty for the current anthropogenic radiative forcing on Earth (IPCC, 2007). In addition, some atmospheric aerosols are pollutants with significant health impacts (Mauderly & Chow, 2008). Finally, aerosols may pose a hazard to some human activities, as was recently demonstrated by the disruption of air traffic by the ash plume from the Eyjafjallajokull volcano in April-May 2010 (Sanderson, 2010). It is then necessary to monitor the spatial and temporal distribution of atmospheric aerosols on scales of few hours and kilometers. Aerosols can be measured in situ. Such measurements allow a full characterization of their composition and particle size distribution, which is needed to evaluate their health impact. Another option is to use

⁎ Corresponding author. Tel.: + 33 169089455; fax: + 33 169087716. E-mail address: [email protected] (F.-M. Bréon). 0034-4257/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2011.06.017

sunphotometers that are directed at the sun (or the moon if the sensitivity is high enough). The resulting measurement provides a direct estimate of the extinction between the top of the atmosphere and the Earth's surface, which is directly related to the aerosol optical depth, with a minor contribution by atmospheric gases. Multi-wavelength measurements provide some information on the particle size distribution (O'Neill et al., 2003). The extinction measurements can be complemented by sky radiance observations, which are inverted to retrieve the full size distribution and some information on the refractive index (Dubovik & King, 2000; Vermeulen et al. 2000; Li et al., 2006). In situ and sunphotometer measurements provide complementary information for characterization of aerosol load, size distribution, shape, and composition. However, being point based, they cannot provide the spatial and temporal coverage required for global monitoring. On the other hand, with their global and repeatable coverage, satellites do have a more global view. Since aerosols selectively scatter solar radiation back to space, satellite instruments that measure solar reflectance are well suited to monitoring global aerosol properties. In fact, the largest aerosol loads such as plumes from volcanic eruptions, biomass burning, or

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

desert dust can be depicted easily on satellite imagery. A quantification of the aerosol load is more difficult however, both because the aerosol contribution to the reflectance is mixed with that of clouds and the surface, and because aerosol scattering depends not only on the load, but also on its size distribution and composition. Several methods have been developed to extract information on atmospheric aerosols from satellite measurements. These methods are briefly described and discussed in the Data section of this paper. Aerosol products derived from satellite observations have already been evaluated against direct sunphotometer measurements (e.g. Kokhanovsky et al., 2007; Levy et al., 2010; Vidot et al. 2008). These are very well suited to a validation exercise because i) both the satellite and the sunphotometer retrieve the same parameter, i.e. the aerosol optical depth; ii) the sunphotometer provides a near-direct measurement of the extinction, which is therefore much more reliable than that from the satellite measurement; iii) there is a sunphotometer network (AERONET) of over 200 ground stations which covers a wide range of surface conditions and aerosol types and iv) this network has a high level of standardization and data are made available with limited restrictions (Holben et al., 1998). Most validation exercises therefore compare satellite estimates to the measurements from sunphotometers. In this paper, we perform a similar exercise. The objective is to use a consistent methodology (same tools, protocols and thresholds) over a large number of satellite aerosol products so that the statistical performance of each can be objectively compared. We make use of the computer facilities of the ICARE Data and Services Center (www.icare. univ-lille1.fr), where many aerosol products are generated or archived. We have been able to access aerosol-load products derived from the measurements of MODIS, MERIS, SEVIRI, POLDER, and CALIOP. These products are evaluated and discussed in the following. Although we acknowledge that other satellite products have been generated, they have not been collected by the ICARE center and therefore not easily accessible for our evaluation. ICARE offers a large selection of satellite data sets related to aerosol, cloud and the water cycle, together with computational facilities. These services significantly lower the load of collecting and archiving large satellite datasets that are necessary when using multi-year observations at high resolution. The data and retrieval algorithms are briefly recalled in Section 2. The matchup method is described in Section 3 and the results in Section 4. Section 5 discusses the results and concludes.

3103

2.1. MODIS MODIS is a multi-spectral sensor that was launched onboard both the NASA Aqua and Terra satellites. Here, we only use products derived from Aqua, as Terra data are not available at ICARE. Over the oceans, the aerosol load and type are estimated from a multi-spectral fit between measured and pre-computed reflectances (Tanré et al., 1997). Because the water column is highly absorbing over most of the solar spectrum, the surface contribution is small compared to that of the atmosphere except close to the specular direction. There is then enough information in the multi-spectral reflectance to estimate both a total aerosol load and a fine mode fraction. We use the official Level2 aerosol product (MYD04_L2), collection 5, generated at 10-km resolution. In the data product, there are many data sets. Our evaluation is based on the following: • • • •

Ångström_Exponent_1_Ocean Effective_Optical_Depth_Average_Ocean Optical_Depth_Small_Average_Ocean Quality_Assurance_Ocean

The retrieval of aerosol load over land is much more difficult than it is over the oceans because the surface contribution is large and variable. It is then difficult to isolate the aerosol signal from the measurement. The MODIS land algorithm uses an empirical relationship that relates surface reflectance in the visible (0.47 and 0.66 um) and middle-infrared (2.1 um) bands (Kaufman et al., 1997). The multi-spectral measurements make it possible to untangle the surface and aerosol respective contributions to the measurements and to derive the aerosol load. The empirical relationship is only valid over vegetated surfaces so that no retrieval is possible over desert areas. The algorithm makes use of a-priori aerosol models that depend on location and season. Our evaluation is based on the following data sets: • • • •

Ångström_Exponent_Land Corrected_Optical_Depth_Land Optical_Depth_Small_Land Quality_Assurance_Land

The MODIS team indicates that there is not enough information to correctly retrieve the fine fraction over land. Which means that the fine model (non-dust) optical depth is not likely to provide information independent of the total optical depth over land. We test this statement in this paper.

2. Data

2.2. MERIS

Table 1 summarizes the sensor characteristics (launch date, viewing swath width, pixel size). Table 2 lists the aerosol product information used in this study together with the references describing the retrieval method. All passive sensors provide column-integrated aerosol measurements while CALIOP observes aerosol vertical profiles in the 0–40 km altitude range.

MERIS is an ESA sensor that was launched onboard Envisat with descending equatorial crossing time of 10:00 local solar time. Like MODIS, it is a multi-spectral sensor, although the spectral coverage is limited to 390–1040 nm, a direct consequence of which is that it cannot use the shortwave-infrared bands used in the MODIS algorithm to estimate the surface contribution over land and no thermal IR bands are

Table 1 Main characteristics of selected satellite sensors. ECT stands for Equatorial crossing time. Instrument/platform

Space agency

POLDER/PARASOL

CNES

MODIS/Aqua

NASA

MERIS/Envisat

ESA

SEVIRI/Meteosat-8 and 9

EUMETSAT

CALIOP/CALIPSO

NASA/CNES

Launch Date ECT* Dec 2004 13:33 UTC May 2002 13:30 UTC March 2002 10:00 UTC Aug 2002 and Dec 2005 June 2006 13:31 UTC

Spatial coverage/swath width Global 1700 km Global 2330 km Global 1150 km Atlantic ocean, Europe-Africa (Geostationary) Global, in lines of 70 m 25° apart

Spatial resolution 5.3 × 6.2 km

2

Period considered in this study Mar. 2005–Jul. 2010

250 m, 500 m, and 1 km

Jan. 2004–Jul 2010

1 km

2004

3 km at nadir

May 2005–Jul 2010

70 m across, 1/3 km along track

June 2006–2008

3104

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

Table 2 Selected aerosol retrieval algorithms and products. AOD: aerosol optical depth, AE: Angström exponent, FM AOD: fine mode AOD, QA: quality assurance. Instrument/ platform

Data provider

Reference

Product name

Used parameters

Product spatial resolution

PARASOL/POLDER

CNES/ICARE

Deuzé et al. (2001) Herman et al. (1997, 2005)

Daily aerosol product over ocean—Level 2 Daily aerosol product over land—Level 2

19 km

MODIS/Aqua

NASA

Ocean: Tanré et al. (1997) Land: Kaufman et al. (1997) Levy et al. (2007)

Aerosol product—Level 2 Collection 005

MERIS/Envisat

ESA

Ocean: Antoine and Morel (1999) Reduced resolution geophysical Land: Santer et al. (1999) product—Level 2

SEVIRI/Meteosat8 and 9

LSCE/ICARE

CALIOP/CALIPSO

NASA

Ocean: Thieuleux et al. (2005) Land: Jolivet et al. (2008) Bernard et al. (2009) Vaughan et al. (2004) Young and Vaughan (2009) Winker et al. (2009)

Ocean: AOD 670 and 865 nm, AE FM AOD 670 nm, FM AE Land: FM AOD 865 nm, FM AE Quality index Ocean: AOD 660 and 860 nm FM AOD 550 nm Land: AOD 670 nm FM AOD 550 nm QA confidence flag Ocean: AOD 865 nm, AE Land: AOD 443 nm, AE QA flags Ocean: AOD 550, AE Land : AOD 630, inverted model

Aerosol optical depth and Angström exponent

5 km-resolution aerosol layers—Level 2

available for cloud screening. The MERIS data are processed by ESA to retrieve various parameters, including the aerosol load, at the original 1 km spatial resolution. Over the ocean, the aerosol load is estimated through a comparison of the measured and pre-computed reflectances, assuming the ocean is black in the near-infrared. The product is the aerosol optical depth at 865 nm “aero_opt_thick”, together with the Ångström exponent “aero_alpha”. We also use the quality flag to retain only the valid retrievals. Over land, the aerosol load is estimated over densely vegetated pixels using an assumption on the surface reflectance spectral signature in the visible. The product is the aerosol optical depth at 443 nm “aero_opt_thick”, together with the Ångström exponent “aero_alpha”. We also use the quality flag to retain only the valid retrievals.

2.3. POLDER/Parasol POLDER is a multi-spectral, multi-directional, and polarizationcapable radiometer. It was launched onboard the ADEOS platforms in the 1990s, but both platforms failed prematurely (Tanré et al., 2001). A similar instrument was launched onboard the PARASOL satellite which is one element of the A-Train. Over the oceans, the retrieval algorithm uses the red and near-infrared bands to estimate the aerosol load. The use of multiple directions which allows a better constrain of the aerosol model compensates for the narrow spectral range. Polarization information is also used which allows a distinction between spherical and non-spherical particles (Herman et al., 2005). We use the aerosol load at 670 nm, the Ångström exponent and the fine mode aerosol products, along with the quality index. Over land, the aerosol retrieval uses polarized reflectance only. The surface contribution to the measurement is assumed based on a semiempirical model (Maignan et al., 2009) and the aerosol contribution is inferred to retrieve the aerosol load and model. The sunlight scattered by the aerosol is highly polarized, but only when particles are small. On the contrary, coarse-mode aerosols polarize very little. As a consequence, POLDER algorithm over land only retrieves the fine mode (FM) optical depth, with no reliable information on the total

Land and Ocean: AOD 500 AOD 865

10 km

1 km (Native)

3 km at nadir (Native)

5 km along track

optical depth. For the validation, we use the FM AOD at 670 nm, the FM Ångström exponent, together with the quality index. 2.4. SEVIRI The SEVIRI radiometer flies onboard the second generation of the Meteosat geostationary satellites above longitude 0°, and provides an image of a portion of the Earth every ≈15 min. The measurements are retrieved by ICARE and processed in near-real time to estimate the aerosol load over the ocean. The measurements from the 635 and 810 nm channels are used in a retrieval algorithm (Thieuleux et al., 2005) to derive the AOD at 550 nm and the Angström exponent. The retrievals are performed at the original SEVIRI resolution (3 km at nadir), and the near real time retrievals can be browsed at www.icare. univ-lille1.fr. Over land, the estimate is derived from a temporal analysis of the signal, assuming that one can get a clear scene image within the 15 previous days. The clear scene is then used to estimate the surface reflectance, which is then applied in a radiative transfer model to estimate the AOD of the scene (Jolivet et al., 2008, Bernard et al., 2009). At the time of writing, the SEVIRI AOD over land product is still under development. Results presented here are preliminary and define a minimum quality that can be expected as potential improvements are already being implemented and tested (Bernard et al., 2011). Both the land and ocean SEVIRI products are not official products of Eumetsat, and other similar “research” algorithms have been developed (e.g. Brindley & Ignatov, 2006; Popp et al., 2007). However, to our knowledge, these algorithms are not used to generate products over long time period and are not available to a wide community. 2.5. CALIOP The CALIOP lidar was launched in April 2006 onboard the CALIPSO satellite and is part of the so-called A-Train since then. Contrary to the instruments mentioned earlier, CALIOP is an active instrument. Processing of the lidar data makes it possible to retrieve the aerosol and cloud vertical distributions, which is not possible with the passive

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

sensors. Although this is of great benefit to the study of aerosol sources and transport, we focus here on the vertically integrated values, i.e. the total AOD as estimated by the algorithm. Another specificity of CALIOP is that it has no cross-track imaging capabilities. The measurements are limited to the satellite subtrack so that an observation coincident with a validation site is less frequent than with the passive instruments. The lidar signal can penetrate thin cloud layers but is fully extinct by dense scattering layers (aerosols or clouds). We therefore exclude cases when the presence of a thick layer prevents a full column observation. In the following, we use the official level-2 “5 km-Resolution Aerosol Layers” product that provides an estimate of the AOD at 532 and 1064 nm. Note that sunlight induces significant noise on the measurements so that nighttime products are of better quality than daytime. However, no sunphotometer data is available during nighttime for validation. Therefore, we cannot use the best quality CALIOP products, which is somewhat unfair for the inter-comparison of various products. In this paper, we use the version-3 products that became available in 2010.

2.6. Sunphotometer (AERONET network) The Aerosol Robotic Network (AERONET) (Holben et al., 1998, 2001) is a well-established network of about 200 semi-permanent groundbased sunphotometers worldwide. They provide standardized high quality aerosol measurements that are used here as reference data. We use the global Level 2.0 data archive (i.e. cloud-screened and qualityassured with up-to-date calibration) up to June 2010. These data were downloaded from the AERONET web page (aeronet.gsfc.nasa.gov). Visible and near-infrared AOD measurements are available with an accuracy of about 0.01–0.02. In addition, we use the AERONET spectral AOD to derive the fine mode AOD (FM AOD) at 550 nm using the spectral deconvolution method of O'Neill et al. (2003). Fig. 1 shows the location of the AERONET sites resulting from the colocation approach (see next section). While a large proportion of the AERONET sites are located in North America and Western Europe, their distribution covers all continents and all aerosol types are sampled allowing a comprehensive evaluation. Note also that many sites are located near the coast, which makes them well suited for a comparison to both land and oceanic retrievals. We here make the hypothesis that the AOD does not change significantly across the land/ocean boundary. We also note that the oceanic retrieval that we use is close to the coast, which adds some difficulty to the retrieval due to higher turbidity and horizontal gradients than over the open ocean. As a consequence, the open ocean performance of AOD estimates is probably better than indicated through the land-based sunphotometer comparison. Finally, we didn't use any selection criterion, such as elevation, to exclude some stations, while it would have been appropriate for some, such as Mauna Loa, where the sunphotometer elevation is much larger than that of its surrounding.

3105

3. Method 3.1. Matching satellite and sunphotometer measurements The method for satellite product evaluation is rather straightforward: we searched for sunphotometer and satellite observations coincident within a spatial and temporal window. There are then two options. One option is to use the temporal average of the sunphotometer measurements within the time window and the spatial average of the satellite data within a given distance from the sunphotometer station. Another option is to use only the closest (in time) sunphotometer observation and the closest (in distance) satellite retrieval. The results shown in the following use the “closest” approach but we compare the results from both approaches in Section 4.6. Note that, for the “average” approach, the mean AOD is computed by weighting individual AOD estimates by their quality index, when available (some products do not include a quality index). Although this approach favours the better quality retrievals, it does make use of lower quality estimates. As shown earlier, the satellite AOD estimates are provided at different wavelengths. Some products provide the estimates at several wavelengths while others provide a single AOD and an Angström exponent. In the former case, we computed an Angström exponent for each interval between the wavelengths that were provided. We then extrapolated the satellite AOD to standard wavelengths, i.e. 500, 670 and 865 nm using  α λ τðλÞ = τðλ0 Þ 0 λ

ð1Þ

where λ is the wavelength, τ is the spectral AOD and α is the Ångström exponent. 3.2. Statistical indicators A question is which statistical indicator to use to quantify the quality of a satellite product. The correlation with the sunphotometer data is an obvious choice. However, a good correlation may be found when a large bias or a slope different to 1 is observed, which is not desirable. Another option is the RMS difference where a small value is sought. However, one problem with the RMS difference is that it favours the small AOD, i.e. RMS difference tends to be smaller over regions that have a small aerosol load than over regions that are highly polluted. To overcome these drawbacks of the correlation and RMS statistical indicators, we favour the “fraction of accurate retrievals” Gfrac and define that a retrieval is “good” when the difference with the sunphotometer data is less than Δτ = 0:05 + 0:15τsunph over land Δτ = 0:03 + 0:05τsunph over the oceans where τsunph is the optical depth measured by the sunphotometer. These thresholds were defined by the MODIS aerosol retrieval team (Remer et al., 2005). They recognized the fact that aerosols are more variable and more difficult to retrieve over land than over the oceans. Although the AOD tends to decrease with wavelength, we use a single value threshold for all spectral bands for simplicity. This has no impact on the primary objective to inter-compare the various retrieval accuracies. 3.3. Time and space window

Fig. 1. Map of AERONET locations used for the validation of satellite products. Land sites are shown in red, ocean sites in blue. Coastal sites (in green) are used for the validation of both Land and Ocean retrievals. There are 110 coastal sites, 202 over land, and 23 oceanic sites. The dashed line shows the SEVIRI limit of visibility, assuming a maximum view zenith angle of 70°.

To analyze which spatial and temporal thresholds should be used, we use MODIS products. For this exercise we used the “closest” evaluation, and we only retained MODIS retrievals with a QAC (Quality Assurance Confidence) flag greater than 0 (ocean) or equal to 3 (land) for reasons that are discussed in the next section. We applied 4 different thresholds for time and 5 for distance. The results are shown in Table 3. One would expect the statistics to improve as the

3106

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

thresholds decrease, and the results are as expected with very few exceptions. The correlations increase, the RMS difference decreases and the Gfrac increases as either threshold decreases. On the other hand, it is obvious that the sample size decreases with the thresholds. There is therefore a compromise to be made. Note that, when looking at global statistics, there are a very large number of samples so that sample size is not an issue. However, it may become one for the analysis of results for individual stations. The tables show that the statistics do not degrade significantly as the spatial threshold increases from 25 to 50 km. On the other hand, there is a strong degradation of statistics from ΔT ≤ 30 min to 30 ≤ ΔT ≤ 60. Based on these considerations, we selected a temporal threshold of 30 min and a spatial threshold of 50 km. All results shown in the following are based on these values, except for CALIOP. Indeed, the lidar has no cross-track scanning capabilities and provides measurements along the subtrack only. As a consequence, the number of observations in coincidence with the sunphotometers is very low, and it is necessary to increase the spatial threshold to get acceptable statistics. For CALIOP comparison, the spatial threshold is therefore set to 150 km.

3.4. Quality flags Both MODIS and POLDER retrievals provide quality indicators within their products. We have therefore attempted to use these indicators and to assess their usefulness. Table 4 shows statistics as a

Table 3 Statistics for MODIS AOD500 retrievals for various thresholds of the space and time window. First line: number of match-ups/correlation, second line: RMS/Gfraction. For each of the statistic indicators, the best value is highlighted in bold characters. Table 3a is for oceanic retrievals, while Table 3b is for the land. Table 3c allows a specific analysis of the matchups that are distant in space and/or time (i.e. between two thresholds rather than lower than a threshold for T3a and T3b). One year of observation has been used for these statistics. (a):OCEAN

Δt ≤ 30 min

Δt ≤ 60 min

Δt ≤ 120 min

Δt ≤ 180 min

d ≤ 25 km

1657/0.785 0.128/56.4 3514/0.798 0.125/51.9 5736/0.743 0.141/48.2 9605/0.654 0.223/43.9 12437/0.602 0.251/41.8

Not done

Not done

Not done

4227/0.799 0.135/51.7 6577/0.717 0.154/47.2 11083/0.630 0.243/43.1 14419/0.591 0.261/41.1

4450/0.747 0.147/49.7 7443/0.697 0.163/45.8 12642/0.629 0.243/41.7 16514/0.592 0.263/39.9

4682/0.750 0.147/48.7 7926/0.694 0.165/44.8 13578/0.628 0.244/41.0 17831/0.588 0.265/39.2

d ≤ 50 km d ≤ 100 km d ≤ 200 km d ≤ 300 km (b): LAND

Δt ≤ 30 min

Δt ≤ 60 min

Δt ≤ 120 min

Δt ≤ 180 min

d ≤ 25 km

5448/0.871 0.137/62.4 6061/0.874 0.142/62.5 6846/0.860 0.152/61.5 7914/0.845 0.164/60.8 8913/0.813 0.174/59.1

5886/0.868 0.139/61.9 6622/0.869 0.145/61.8 7533/0.861 0.154/60.9 8762/0.841 0.167/60.1 9904/0.811 0.178/59.0

6232/0.870 0.143/61.0 7115/0.869 0.149/60.8 8191/0.860 0.158/59.8 9620/0.831 0.174/58.9 10922/0.802 0.185/57.9

6384/0.867 0.144/60.7 7354/0.857 0.156/60.3 8522/0.848 0.165/59.3 10067/0.821 0.179/58.3 11470/0.793 0.189/57.3

d ≤ 50 km d ≤ 100 km d ≤ 200 km d ≤ 300 km (c): LAND

Δt ≤ 30 min

30 ≤ Δt ≤ 60

60 ≤ Δt ≤ 120

120 ≤ Δt ≤ 180

d ≤ 25 km

5448/0.871 0.137/62.4 621/0.932 0.134/63.6 775/0.787 0.217/54.6 1087/0.760 0.222/56.4 1014/0.597 0.244/52.0

438/0.824 0.170/55.7 126/0.757 0.199/49.2 131/0.942 0.159/58.8 161/0.606 0.275/48.4 150/0.764 0.283/44.7

346/0.885 0.190/45.7 145/0.814 0.199/51.7 162/0.805 0.215/48.8 197/0.306 0.330/43.1 161/0.705 0.266/45.3

152/0.762 0.200/47.4 86/0.671 0.403/36.0 92/0.692 0.273/44.6 118/0.604 0.205/44.1 106/0.593 0.197/47.2

25 ≤ d ≤ 50 50 ≤ d ≤ 100 100 ≤ d ≤ 200 200 ≤ d ≤ 300

function of the quality indicator for MODIS while Table 5 does the same for POLDER. MODIS retrievals over land are clearly much better when QAC = 3 than for other values of QAC. It is then highly advocated to use only those retrievals, as recommended in the collection-5 ATBD (Remer et al., 2009) and in Levy et al. (2010). Over the oceans, there is no clear trend with the quality index. The results are very poor for QAC = 0, but this concerns only a very small fraction of the retrievals. We will therefore use cases with QAC ≥ 1. As for POLDER, the quality indicator may take a larger number of values (from 0 to 100 that is converted here from 0 to 1). Over land, cases with Q ≥ 0.6 are much more numerous than for lower values. When the quality index increases, the correlation and Gfrac tend to increase, but the RMS does not decrease significantly. It is likely that this is due to different aerosol loads in the different classes: retrieval is easier when there is some aerosol signal, i.e. for a significant aerosol load. These results suggest the use of a threshold of 0.5 for the quality index, and this is what is done in the following. Over the oceans, there is no significant trend when the quality index is greater than 0.2. We will therefore subsequently use this value as a threshold. There is no quantitative quality index for MERIS, but confidence flags and flags relevant to product interpretation are given together with the AOD estimates (Sotis, 2007). We therefore computed statistics only in favourable conditions (“High Quality”), i.e.: over ocean: far from the glint and for case-1 waters (non turbid); over land: pixels with Dark Dense Vegetation (DDV). For SEVIRI over land, retrievals are attempted whenever a clear sky pixel is identified. However, a binary flag defines the retrieval as useful (confident) or non-useful (low confidence) depending on local variability of retrieved AOT, confidence in clear scene identification and sun-sensor geometries. Only retrievals defined as useful over land were retained here for evaluation. 4. Results Figs. 2–5 shows density histograms of the various AOD estimates against AERONET. The wavelength is 670 nm, but we have made similar plots for other channels as well. We have selected this channel because it is an average between the standard wavelengths of the various sensors (from 440 to 865). Table 6 provides the statistical results for two other channels as well (500 and 865 nm). There are four subplots that correspond to ocean and land-based retrievals, and total and fine mode AOD. Only MODIS and POLDER provide a fine mode AOD, whereas POLDER does not estimate a total AOD over land, which explains the varying number of scatter plots per figure. Note also that CALIOP algorithm is the same over land and ocean, and that it is less affected by the surface than the passive sensors. We therefore did not distinguish between land and ocean for CALIOP. These figures are generated with the quality indices and the spatio-temporal thresholds discussed in the previous section. 4.1. Total AOD over the ocean The total AOD over the ocean is shown in Fig. 2. It is clear that neither the MERIS nor CALIPSO daytime AOD products provide quantitative information. The MERIS AODs are large and there is no

Table 4 Same statistical indicators as in Table 3, but as a function of the quality indicator in the MODIS product. The statistics are based on the optical depth at 500 nm. Five years of data have been used for these statistics.

QAC = 0 QAC = 1 QAC = 2 QAC = 3

Land

Ocean

20871/0.808/0.202/45.8 17403/0.821/0.191/49.1 16120/0.843/0.174/53.0 23047/0.903/0.126/67.9

260/0.701/0.587/44.6 19749/0.792/0.116/53.5 0 5510/0.829/0.151/42.5

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

3107

Table 5 Same as Table 4, but for the POLDER/Parasol product. Over land, the statistics are based on the fine mode optical depth at 550 nm.

0 ≤ Q ≤ 0.2 0.2 ≤ Q ≤ 0.4 0.4 ≤ Q ≤ 0.6 0.6 ≤ Q ≤ 0.8 0.8 ≤ Q ≤ 1.0

Land

Ocean

1567/0.112/0.154/52.5 1736/0.272/0.119/55.5 5228/0.370/0.114/59.5 17766/0.678/0.114/63.7 18846/0.882/0.121/71.3

1180/0.508/0.307/29.2 952/0.915/0.110/36.9 2764/0.875/0.115/45.6 6410/0.879/0.105/50.3 1222/0.886/0.106/51.6

correlation with AERONET. The MERIS algorithm is designed primarily for atmospheric correction of ocean color retrievals. The aerosol correction probably includes contributions from other perturbations, such as clouds, glint or sediments so that the aerosol products should not be used for aerosol studies (Ramon & Santer, 2008; Vidot et al., 2008). Similarly, there is poor agreement between CALIPSO AOD and AERONET. One may recall that the comparison with the other sensors is somewhat unfair, firstly because the data used here do not include nighttime retrievals, when CALIPSO performs best, and also because there is, on average, a much greater distance between the spaceborne and surface observations than with the other sensors. Nevertheless, solely based on the AERONET reference, it seems that the CALIPSO daytime AOD product provides little information on the total column AOD. At the other extreme, POLDER and MODIS retrievals show a rather high correlation with the sunphotometer data (0.91 and 0.85 respectively). Although POLDER correlation is the highest, MODIS Gfrac is significantly better (57% vs 45%). This is because of a high bias in POLDER retrievals that is apparent for the clean atmosphere (lower left corner of the far left plot in Fig. 2). As a consequence, and because the “clean” case is the most frequent, a large fraction of POLDER retrievals fall outside the valid retrieval limits. Also note that POLDER seems to retrieve more high AODs than MODIS does (at least with the quality index criteria), which favours a high correlation. Finally, SEVIRI retrievals are not as good as those from POLDER or MODIS, but they are nevertheless highly correlated with the sunphotometer data (0.72). There is a clear slope in the comparison (large AODs are slightly too large) which may indicate a bias in the aerosol phase function that is used for the retrievals. As a consequence, and although the correlation is high, Gfrac is only 34%. There seems to be a larger proportion of high AODs with SEVIRI than with POLDER or MODIS, but this may result from the geostationary satellite field of view that is centered on locations affected by the Saharan dust transport.

Fig. 3. Same as Fig. 2 but for the Fine Mode AOD at 550 nm. Only POLDER and MODIS provide such estimate.

most often found with small Ångström exponent, which indicates the presence of dust. The proportion of fine mode in the dust is small, which explains the lower values. The performance of POLDER and MODIS retrievals are similar with no significant bias, a Gfrac around 53%, and a correlation around 0.76. The performances in terms of fine mode AOD are nevertheless lower than those for the total AOD which indicates the difficulty in deriving size information from the spectral or directional reflectance. There may also be some ambiguity between the definition of “fine mode” between the sunphotometer and satellite products (Kleidman et al., 2005). 4.3. Total AOD over land Three total AOD products derived from MODIS, MERIS and SEVIRI are evaluated here. It is clear from the scatter plots that only MODIS has the most value over land. It shows a rather high (0.86) correlation with the ground-based measurements. Although there is a slight positive bias (0.02), the density plot indicates a small underestimate for the clean cases. Note that the correlation is slightly better than that found over the oceans, although the density histogram appears more dispersed. This is because a few points with large AOD, off scale in Fig. 4, tend to pull the correlation. Gfrac is also significantly better than it is over the oceans, but this is only due to less stringent requirements over land. As over the oceans, MERIS retrievals are much too large and result in large positive biases. On the other hand, there is some value in SEVIRI retrievals, with a correlation of 0.63 and a Gfrac of 52%. As over the ocean, the SEVIRI product is not as good as that from MODIS, but its high temporal resolution may be useful for some applications and therefore compensate somewhat its higher noise.

4.2. Fine mode AOD over the ocean 4.4. Fine mode AOD over land POLDER and MODIS provide an estimate of the fine mode AOD. As expected, the range of values is smaller than for the total AOD. An analysis distinguishing between several Ångström exponent ranges (not shown) demonstrates that the large optical depth values are

POLDER and MODIS provide a fine mode AOD estimate over land (Fig. 5). They both show a significant correlation with the sunphotometer measurements (0.84 and 0.75). For this particular case, all statistical

Fig. 2. Density histograms of the spaceborne estimates of the AOD670 (Y-axis) against sunphotometer measurements (X-axis) for oceanic retrievals. The dashed lines indicate the limits for a “good” retrieval which are used to estimate Gfrac. The numerical values are reported in Table 6. Calipso retrievals are for both land and oceans. The color scale of the density histogram is shown in Fig. 3.

3108

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

Fig. 4. Same as Fig. 2 but for land surfaces.

Based on the aforementioned results, it appears that the best products, in terms of accuracy, are either POLDER or MODIS. The SEVIRI product over the oceans appears relatively good, and it has some added value due to its much higher temporal resolution. We therefore concentrate on these aerosol products and analyze whether there are some spatial variations in their performances. Fig. 6 shows several maps, for each sensor and product type. The color of each dot indicates the proportion of good retrievals, Gfrac, for the corresponding AERONET station. The first two lines are for the oceanic retrievals of POLDER (left) and MODIS (right). Both sensors show good performances over the open oceans (island stations), over the coastlines of Europe and the United States. On the other hand, the performances are lower for the coastal stations of Asia. The performances of the numerous stations located on the coastlines of the Persian Gulf are lower than those located on the coast of the Mediterranean Sea but, in both cases, MODIS performs better than POLDER. SEVIRI results (lower left) can be compared to those of MODIS and POLDER (second line). The SEVIRI performances are systematically poorer than those of the two polar-orbiting sensors. A careful analysis shows that SEVIRI performances are very unsatisfactory over two stations (black dots in the lower-left subplot of Fig. 6) located in Chibolton, UK, and Buenos Aires, Argentina. Further analysis indicates that the retrievals for these two stations are made over very turbid waters. Turbid/coastal waters have a reflectance that is much larger than that of clear waters (Li et al., 2003), which has a direct impact on the

4.6. Impact of spatio-temporal averaging We now analyze whether the statistics improve when performing some spatio-temporal averaging. One certainly expects that averaging will reduce the random error so that the statistics should improve. On the other hand, the averaging makes use of measurements made further away in space (satellite) and time (sunphotometer) than those used for the statistics shown so far so that some degradation can

Table 6 Statistical results for the various sensors. For each sensor and wavelength, one provides the number of matchups, the correlation, the RMS and Gfrac. For each wavelength, surface type and statistical indicator, the best value is highlighted in bold. For a given sensor, the number of matchups is variable because some Aeronet sites do not provide measurements at all standard wavelengths.

POLDER MODIS MERIS SEVIRI CALIOP POLDER MODIS MERIS

Fig. 5. Same as Fig. 2 but for the fine mode AOD at 550 nm for Land Surfaces.

SEVIRI

500 nm

670 nm

870 nm

Fine mode 550 nm

10302/0.886 0.105/44.1 22184/0.829 0.118/48.0 5119/0.087 1.079/ 4.1 42811/0.737 0.235/20.7 5767/0.376 0.404/23.5

14221/0.910 0.087/45.4 30375/0.849 0.094/56.6 6000/0.109 0.834/ 3.8 72498/0.771 0.158/31.6 7647/0.478 0.276/31.7

14154/0.916 0.079/50.4 30259/0.845 0.085/63.1 5964/0.119 0.729/ 4.3 75669/0.730 0.142/40.0 7550/0.501 0.275/32.9

14184/0.773 0.077/51.8 30327/0.746 0.080/53.1

−

−

−

21603/0.904 0.125/68.3 5921/0.341 0.340/32.8 83504/0.626 0.202/39.8

27155/0.864 0.115/69.2 7410/0.309 0.307/26.4 149186/0.629 0.165/52.3

6787/0.814 0.108/70.0 7368/0.263 0.293/25.5 147967/0.559 0.161/61.5

−

OCEAN

4.5. Geographical distribution

aerosol load estimate that assumes a dark surface. The SEVIRI algorithm lacks a turbid water test and flag which has a direct consequence on these particular sites, and probably also on others. Many remote sensing algorithm use a threshold on the water turbidity. This is the case for POLDER that therefore rejected both Chibolton and Buenos Aires points. The MODIS algorithm does include a turbidity flag. There is no valid estimate in the vicinity of Chibolton and a few around Buenos Aires that have a quality lower than average. Over land, MODIS AOD performances appear rather homogeneous (lower right subplot) with no clear regional trend. On the other hand, the fine mode AOD results are contrasted, with the best performances over the Americas, North and South, lower quality results over Western Europe, and rather poor performances over the Sahel, India and South-East Asia. These are the regions where null fine mode AOD is provided in the MODIS product. POLDER fine mode AOD appears of better quality than MODIS almost everywhere, and does not show any significant regional trend.

− − 36096/0.840 0.113/67.6 27166/0.753 0.148/55.7 − −

LAND

indicators indicate that POLDER retrievals are better than those of MODIS. On the other hand, it appears that both products underestimate the largest fine mode AOD (off scale in Fig. 5). Further research shows that MODIS statistics are very much degraded by cases when its fine mode AOD is at zero. If we remove those cases when computing the statistics, the correlation increases to 0.85 and Gfrac is 74%. These values make the performance of MODIS product better than that of POLDER, but the RMS of the former remains slightly larger than that of the latter. Clearly, values of fine mode AOD at zero cannot be trusted and their elimination very much increases MODIS statistical performance.

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

3109

Fig. 6. Fraction of good retrievals Gfrac for different sensors and parameters, and for ocean and land retrievals. The right column is MODIS retrievals. The left columns are the corresponding POLDER retrievals, except for the lower subplot that is SEVIRI over the oceans (and can be compared to the second line of plots for POLDER and MODIS).

therefore be expected. Table 7 shows the results for the various sensors where the first line corresponds to the “closest” approach (as for all results shown so far) and the second line is for the “average”

approach. All statistical indicators, with the exception of MERIS over land, improve through averaging, which clearly demonstrates that there are some random errors in the estimates. On the other hand, the

3110

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111

Table 7 Statistical results for the various sensors. For each sensor, we show the results for the “closest” (first line) and “average” (second line) methods. The statistics are for the total optical depth at 670 nm, except for POLDER over land that shows the fine mode AOD results. CALIOP results are for land and ocean combined. Bold highlights the only statistics that do not improve through averaging. Ocean POLDER MODIS MERIS SEVIRI CALIOP

Land

14221/0.910/0.087/45.4 36155/0.840/0.113/67.6 14221/0.916/0.083/47.1 36155/0.859/0.109/72.6 30375/0.849/0.094/56.6 27155/0.864/0.115/69.2 30375/0.857/0.090/57.4 27155/0.890/0.096/76.2 6000/0.109/0.834/ 3.8 7410/0.309/0.307/26.4 6000/0.198/0.521/ 7.3 7410/0.283/0.293/23.7 72498/0.771/0.158/31.6 149186/0.629/0.165/52.3 72498/0.795/0.138/39.8 149186/0.674/0.153/59.8 7647/0.478/0.276/31.7 7647/0.587/0.213/34.1

improvement is not striking. The retrievals from MERIS over ocean improve significantly through averaging, but still remain poor. Both POLDER and MODIS show a significant improvement over land, but not over the ocean, which indicates that the random errors are more important over land. Except for MERIS, the statistical indicators are consistent in showing that the error decreases only slightly through spatial and temporal averaging. Random errors decrease through averaging while systematic errors do not. Since statistically random errors are decreased by averaging, while systematic biases are not, our results indicate that the systematic errors are dominating the error budget of satellite products. 5. Summary and conclusion In this paper, we provide an evaluation of current operational aerosol products from MODIS, POLDER, SEVIRI, CALIOP and MERIS, over both land and ocean. We performed tests to define an optimal comparison procedure. The analysis of the results with various time-space match-up windows indicates that a 30 min/50 km-radius window is appropriate for satellite/ground-based comparisons over land. Although these thresholds were derived only from MODIS products, we assume they can also apply to other sensors. We used similar thresholds for oceanic retrievals, although it is clear that the spatial scale of aerosol plumes over the open oceans is likely to be much larger than over coastal areas where our evaluation was conducted. We showed that POLDER and MODIS retrieval agreement with sunphotometer data over land depends very much on the use of the quality indices, but that there is no such strong link for oceanic retrievals. We therefore suggest using only the highest quality retrievals over land (see text for details), and removing only the lowest quality indices over the oceans. Among the datasets that we tested, MODIS and POLDER provide the most accurate information about aerosol load. Over the oceans, these two products have a similar accuracy and their performance ranking depends on the statistical indicator. POLDER tends to show a better correlation with AERONET, but with a small bias that has a negative impact on the proportion of good retrievals Gfrac. The aerosol-load estimate derived from SEVIRI is not as good as that from POLDER or MODIS, but nevertheless correlates well with the sunphotometer data. In addition, SEVIRI provides a temporal resolution that is far greater than that of the other spaceborne sensors, which may be needed for many applications. SEVIRI should therefore be considered for aerosol monitoring over the ocean, within its field of view. In addition, its statistics could probably improve with a rigorous elimination of turbid water cases. Over land surfaces, POLDER can only provide an estimate of the fine mode AOD, but this estimate is clearly better than that from MODIS or any other sensor. As for the total AOD, MODIS provides the

best estimate, with a correlation against AERONET of 0.86, and 69% of the retrievals considered as “good”. This result suggests a combined product from POLDER and MODIS, with the total AOD from MODIS and the fine mode from POLDER. On the other hand, the MERIS aerosol products, both over land and the oceans, are of doubtful value with estimates much larger than the sunphotometer measurements, and poorly correlated. Similarly, SEVIRI estimates over land show a noise which is larger than that of MODIS, but the satellite product is significantly correlated (0.63) with the sunphotometer measurements. There is therefore value in the product, in particular for applications that require high temporal frequency. The CALIOP-derived AOD estimates show little resemblance to the sunphotometer data. The comparison with the passive sensor is somewhat unfair, firstly because one does not use the nighttime estimates, which are expected to be of better quality because of the reduced measurement noise, and also because of the typically large distance between the satellite data and the ground-based observation, a result of the lack of swath on the lidar. Nevertheless, our results suggest the CALIOP AOD product should not be used for quantitative studies. Since the Aqua, PARASOL and CALIPSO satellites fly in close formation, one may rather use CALIOP to assess the vertical distribution of aerosols and use it in synergy with the total AOD from either of the passive sensors. Acknowledgements It is fair to state that the main author of this research has significant links with the POLDER/Parasol research group. Although the study presented in the paper tried to apply consistent and fair evaluation procedures, unconscious biases are always possible. We acknowledge the support of the European Commission through the GEOmon (Global Earth Observation and Monitoring) Integrated Project under the 6th Framework Program (contract number FP6-2005-Global-4-036677). We thank CNES for providing the Level-1 POLDER/PARASOL data (Level-2 and Level-3 products and browse images courtesy of ICARE/LOA/LSCE), NASA for providing the Aqua/MODIS data (browse images courtesy of ICARE), NASA and CNES for providing the CALIOP/CALIPSO data (browse images courtesy of ICARE), EUMETSAT/ EUMETCAST and SATMOS for providing the MSG/SEVIRI data (products and browse images courtesy of ICARE/LSCE), and ESA for providing the MERIS data. We also thank the AERONET project at NASA/GSFC for providing the ground-based aerosol data. We thank Norm O'Neill, University of Sherbrooke, for providing his code to compute the fine and coarse mode aerosol optical depth from AERONET data. And last but not least, we acknowledge and thank three anonymous reviewers who provide useful comments and recommendations. References Antoine, D., & Morel, A. (1999). A multiple scattering algorithm for atmospheric correction of remotely sensed ocean colour (MERIS instrument): principle and implementation for atmospheres carrying various aerosols including absorbing ones. International Journal of Remote Sensing, 20(9), 1875–1916. Bernard, E., Moulin, C., Ramon, D., Jolivet, D., Riedi, J., & Nicolas, J. M. (2011). Validation of an AOT product over land at the 0.6 μm channel of the SEVIRI sensor onboard MSG. Atmospheric Measurement Techniques Discussions, 4, 3147–3198, doi:10.5194/amtd-43147-2011. Bernard, E., Ramon, D., Jolivet, D., Moulin, C., Riedi, J., Deschamps, P. -Y., Nicolas, J. -M., & Hagolle, O. (2009). Aerosol retrieval over land in the 635 nm channel of MSG/SEVIRI sensor: a hourly and daily AOT product over Europe. Proceedings of the EUMETSAT Meteorological Satellite Conference, Bath, United Kingdom. Brindley, H. E., & Ignatov, A. (2006). Retrieval of mineral aerosol optical depth and size information from Meteosat Second Generation SEVIRI solar reflectance bands. Remote Sensing of Environment, 102, 233–363. Deuzé, J. L., Bréon, F. M., Devaux, C., Goloub, P., Herman, M., Lafrance, B., Maignan, F., Marchand, A., Nadal, F., Perry, G., & Tanré, D. (2001). Remote sensing of aerosols over land surfaces from POLDER-ADEOS-1 polarized measurements. Journal of Geophysical Research, 106(D5), 4913–4926.

F.-M. Bréon et al. / Remote Sensing of Environment 115 (2011) 3102–3111 Dubovik, O., & King, M. D. (2000). A flexible inversion algorithm for retrieval of aerosol optical properties from Sun and sky radiance measurements. Journal of Geophysical Research - Atmosphere, 105(D16), 20673–20696. Herman, M., Deuzé, J. L., Devaux, C., Goloub, P., Bréon, F. M., & Tanré, D. (1997). Remote sensing of aerosols over land surfaces, including polarisation measurements. Application to Polder measurements. Journal of Geophysical Research, 102, 17039–17049. Herman, M., Deuzé, J. L., Marchand, A., Roger, B., & Lallart, P. (2005). Aerosol remote sensing from POLDER/ADEOS over the ocean: improved retrieval using a nonspherical particle model. Journal of Geophysical Research, 110. Holben, B. N., Eck, T. F., Slutsker, I., Tanre, D., Buis, J. P., Setzer, A., Vermote, E., Reagan, J. A., Kaufman, Y., Nakajima, T., Lavenu, F., Jankowiak, I., & Smirnov, A. (1998). AERONET — a federated instrument network and data archive for aerosol characterization. Remote Sensing of Environment, 66, 1–16. Holben, B. N., Tanré, D., Smirnov, A., Eck, T. F., Slutsker, I., Abuhassan, N., et al. (2001). An emerging ground-based aerosol climatology: aerosol optical depth from AERONET. Journal of Geophysical Research, 106, 12 067–12 097. IPCC (2007). Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. In S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M.Tignor, H.L. Miller (Eds.), Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Jolivet, D., Ramon, D., Bernard, E., Deschamps, P. -Y., Riedi, J., Nicolas, J. -M., & Hagolle, O. (2008). Aerosol monitoring over land using MSG/SEVIRI. Proceedings of the EUMETSAT Meteorological Satellite Conference, Darmstadt, Germany. Kaufman, Y. J., Tanré, D., Remer, L., Vermote, E., & Holben, B. N. (1997). Operational remote sensing of tropospheric aerosols over the land from EOS-MODIS. Journal of Geophysical Research, 102, 17051–17068. Kleidman, R. G., O'Neill, N. T., Remer, L. A., Kaufman, Y. J., Eck, T. F., Tanre, D., Dubovik, O., & Holben, B. N. (2005). Comparison of Moderate Resolution Imaging Spectroradiometer (MODIS) and Aerosol Robotic Network (AERONET) remote-sensing retrievals of aerosol fine mode fraction over ocean. Journal of Geophysical Research Atmospheres, 110, D22205, doi:10.1029/2005JD005760. Kokhanovsky, A. A., Breon, F. M., Cacciari, A., Carboni, E., Diner, D., Di Nicolantonio, W., et al. (2007). Aerosol remote sensing over land: a comparison of satellite retrievals using different algorithms and instruments. Atmospheric Research, 85(3-4), 372–394 (15 authors in alphabetic order). Levy, R. C., Remer, L. A., Kleidman, R. G., Mattoo, S., Ichoku, C., Kahn, R., & Eck, T. F. (2010). Global evaluation of the Collection 5 MODIS dark-target aerosol products over land. Atmospheric Chemistry and Physics Discussion, 10, 14815–14873, doi:10.5194/acpd-1014815. Levy, R. C., Remer, L. A., Mattoo, S., Vermote, E. F., & Kaufman, Y. J. (2007). Secondgeneration operational algorithm: Retrieval of aerosol properties over land from inversion of Moderate Resolution Imaging Spectroradiometer spectral reflectance. Journal of Geophysical Research, 112, D13211. Li, Z., Goloub, P., Devaux, C., Gu, X., Deuze, J. L., Qiao, Y., & Zhao, F. (2006). Retrieval of aerosol optical and physical properties from ground-based spectral, multi-angular, and polarized sun-photometer measurements. Remote Sensing of Environment, 101 (4), 519–533, doi:10.1016/j.rse.2006.01.012. Li, R. -R., Kaufman, Y. J., Gao, B. -C., & Davis, C. O. (2003). Remote sensing of suspended sediments and shallow coastal waters. IEEE TGARS, 41(3), 559–566.

3111

Maignan, F., Bréon, F. -M., Fédèle, E., & Bouvier, M. (2009). Polarized reflectances of natural surfaces: spaceborne measurements and analytical modeling. Remote Sensing of Environment, 113, 2642–2650. Mauderly, J. L., & Chow, J. C. (2008). Health effects of organic aerosols. Inhalation Toxicology, 20(3), 257–288, doi:10.1080/08958370701866008. O'Neill, N. T., Eck, T. F., Smirnov, A., Holben, B. N., & Thulasiraman, S. (2003). Spectral discrimination of coarse and fine mode optical depth. Journal of Geophysical Research, 108(D17), 4559, doi:10.1029/2002JD002975. Popp, C., Hauser, A., Foppa, N., & Wunderle, S. (2007). Remote sensing of aerosol optical depth over central Europe from MSG-SEVIRI data and accuracy assessment with ground-based AERONET measurements. Journal of Geophysical Research, 112, D24S11, doi:10.1029/2007JD008423. Ramon, D., & Santer, R. (2008). Validation of the MERIS aerosol product over land. Proceedings of the 2nd MERIS/AATSR workshop, ESA/ESRIN Frascati, Italy. Remer, L. A., Kaufman, Y. J., Tanré, D., Mattoo, S., Chu, D. A., Martins, J. V., et al. (2005). The MODIS Aerosol algorithm, products, and validation. Journal of the Atmospheric Sciences, 62, 947–973. Remer, L. A., Tanré, D., Kaufman, Y. J., Levy, R. C., & Mattoo, S. (2009). Algorithm for remote sensing of tropospheric aerosol from MODIS: collection 005: Revision 2. Sanderson, K. (2010). Questions fly over ash-cloud models. Nature, 464, 1253, doi:10.1038/ 4641253a. Santer, R., Carrere, V., Dubuisson, P., & Roger, J. C. (1999). Atmospheric correction over land for MERIS. International Journal of Remote Sensing, MERIS special issue, 20, 1819–1840. Sotis, G. (2007). Envisat-1 product specifications, Volume 11: MERIS product specifications. Issue 5, Revision B. PO-RS-MDA-GS-2009, ESA/ESRIN, 102–105. Tanré, D., Bréon, F. -M., Deuzé, J. L., Herman, M., Goloub, Ph., Nadal, F., & Marchand, A. (2001). Global observation of anthropogenic aerosols from satellite. Geophysical Research Letters, 28(24), 4555–4558. Tanré, D., Kaufman, Y. J., Herman, M., & Mattoo, S. (1997). Remote sensing of aerosols properties over oceans using the MODIS/EOS spectral radiances. Journal of Geophysical Research, 102, 16971–16988. Thieuleux, F., Moulin, C., Bréon, F. M., Maignan, F., Poitou, J., & Tanré, D. (2005). Remote sensing of aerosols over the oceans using MSG/SEVIRI imagery. Annales Geophysicae, 23(12), 3561–3568. Vaughan, M. A., Young, S. A., Winker, D. M., Powell, K. A., Omar, A. H., Liu, Z., Hu, Y., & Hostetler, C. A. (2004). Fully automated analysis of space-based lidar data: an overview of the CALIPSO retrieval algorithms and data products. Proceedings of SPIE, 5575, 16–30. Vermeulen, A., Devaux, C., & Herman, M. (2000). Retrieval of the scattering and microphysical properties of aerosols from ground-based optical measurements including polarization. I. Method. Applied Optics, 39(33), 6207–6220. Vidot, J., Santer, R., & Aznay, O. (2008). Evaluation of the MERIS aerosol product over land with AERONET. Atmospheric Chemistry and Physics, 8(24), 7603–7617. Winker, D. M., Vaughan, M. A., Omar, A., Hu, Y. X., Powell, K. A., Liu, Z. Y., Hunt, W. H., & Young, S. A. (2009). Overview of the CALIPSO mission and CALIOP data processing algorithms. Journal of Atmospheric and Oceanic Technology, 26(11), 2310–2323. Young, S. A., & Vaughan, M. A. (2009). The retrieval of profiles of particulate extinction from cloud-aerosol lidar infrared pathfinder satellite Observations (CALIPSO) data: algorithm description. Journal of Atmospheric and Oceanic Technology., 26(6), 1105–1119.