Multi-sensor satellite remote sensing of dust aerosols over North Africa during GERBILS

Authors


Abstract

In this paper we provide an overview of various satellite products over the Sahara Desert that were available during the GERBILS field campaign. Our results indicate that all mid-visible satellite aerosol optical depth (AOD) products match well with AERONET retrievals. For low AOD (AOD < 1), the satellite AODs compare well with aircraft AOD values but they tend to underestimate at high AOD values. We then assessed the satellite products in 0.5 × 0.5 degree grids for the entire study region (10–30°N and 20°W–10°E). If we use a multi-angle imaging spectroradiometer (MISR) as a benchmark for AOD retrievals over bright targets, the estimated AOD derived from the ozone-monitoring instrument aerosol index–MISR relationship performs best when compared with MISR for the entire study region. Although differences exist among satellite products, the advancement in satellite retrieval techniques now provide AOD retrievals over bright targets such as deserts, which are useful for numerical modeling simulation comparisons and other studies. Furthermore, the in situ information from aircraft and the ground continue to provide valuable information for validating satellite products and for assessing their strengths and weaknesses. Copyright © 2011 Royal Meteorological Society and British Crown Copyright, theMet Office

1. Introduction

The Sahara in North Africa, which covers over 9 million km2, produces more than 50% of global dust aerosols (Goudie and Middleton, 2001). Although dust aerosols are produced throughout the year in Africa, there are distinct seasonal cycles, with high concentrations of dust produced during summer months, although the Bodele has peak values during winter and spring months. Dust aerosols have a varying range of effects and consequences since they change the radiative balance of the Earth's atmosphere system in both the solar and terrestrial regions of the electromagnetic spectrum (Haywood et al., 2001; Patadia et al., 2009; Yang et al., 2009). In the solar region dust scatters and absorbs sunlight, leading to more radiation being scattered to space, and a reduction in sunlight at the surface, leading to a negative radiative forcing and an associated cooling. In the terrestrial region, mineral dust absorbs outgoing terrestrial radiation and re-emits radiation at a lower temperature both to space and back to the surface of the Earth, leading to a positive radiative forcing and an associated warming (Yang et al., 2009). The radiative balance and the resulting impact on surface temperatures are a complex result of how absorbing the particles are: their shape, composition, size distribution and surface reflectance, absorption and emissivity properties. All of these variables are spatially and temporally dependent and they are a function of location and proximity to sources and various atmospheric processes.

Desert regions are some of the harshest environments for making routine measurements from the ground, partly owing to their distance from human resources. However, one of the great successes of the last 10–15 years of aerosol measurements is due in part to the AErosol RObotic NETwork (AERONET) program, where routine retrievals of aerosol optical depth (AOD), which is a unitless columnar measure of aerosol loading, are made at multiple wavelengths at various locations throughout the world, including North Africa (Holben et al., 2001). Currently there are several hundred AERONET measurements worldwide and North Africa has several sites. Sub-hourly values of AOD coupled with daily measurements of aerosol size, single scattering albedos and others are routinely made available to the world-wide community (Dubovik and King, 2000). However, these measurements only represent column integrated values at specific locations. Complementary vertical information over larger spatial scales can be obtained from airborne measurements made during field experiments (Haywood et al., 2011). Since the in situ measurements cannot cover large spatial areas, space-borne measurements from satellites are increasingly becoming important to map the spatial distribution of aerosols and their properties and to assess impacts (Table I). However, ground and airborne measurements are crucial to interpret and validate the satellite measurements (e.g. Johnson et al., 2011).

Table I. Satellite datasets and products over North Africa.
SatelliteSensorKey wavelengths used (nm)Product nameCoverageProduct spatial resolution (km2)Key references
Aqua/TerraMODIS470, 670, 212Collection 5Daily10 × 10Remer (2005)
AquaMODIS412, 490, 670Deep BlueNear daily10 × 10Hsu et al. (2006)
TerraMISR447,558,672, 867 (9 cameras)MIL2ASAEWeekly17.6 × 17.6Kahn et al. (2005)
AuraOMI354, 388OMAERUVDaily13 × 24Torres et al. (2007)
AuraOMI342-483 (14 bands)OMIAER0 (not used)Daily13 × 24Curier et al. (2008)
AquaMISR-OMI354, 388EAODDaily13 × 24Christopher et al. (2008)
PARASOLPOLDER856, 670, 443AOD (not used)Daily20 × 20Herman et al. (2005)
METEOSATSEVIRI8, 11, 12 µmDust Index15 min10 × 10Brindley and Russell (2009)

In June 2007, the GERBILS campaign was conducted to study dust aerosols in West Africa (Haywood et al., 2011). The Met Office BAE-146 aircraft made 10 flights (Table II) primarily between Niamey and Noukachott (Fig. 1) and performed various measurements of aerosol properties. One of the goals of this paper is to assess the various satellite products over North Africa that are useful for verifying model forecasts (Johnson et al., 2011). The objectives of this paper are twofold: to provide an overview of the various satellite products over the Saharan desert during the GERBILS field campaign; and to inter-compare these satellite products and assess the satellite-retrieved AODs with aircraft and ground-based measurement values when applicable.

Figure 1.

Flight tracks during GERBILS (also see Table I for description). Superimposed on these tracks are AERONET locations where the comparisons between satellite retrievals and AERONET AODs were conducted. Also shown are the white sky/clear sky albedos for June 2007. The figure also indicates locations of the aircraft profiles for each flight (as squares) from which AODs were calculated.

Table II. GERBILS flights and descriptions.
FlightDateFlight description
B29418 JuneDust over ocean
B29519 JuneMISR overpass near Noukchaott
B29621 JuneHuge dust storm over southern Mali and Senegal
B29722 JuneDust over ocean
B29822 JuneTransit flight, late in the evening
B29924 JuneStandard flight mostly along 18°N
B30025 JuneStandard route along 18°N (case study for this paper)
B30127 JuneLong low-altitude run along 18°N with MISR overpass
B30228 JuneLong low-altitude run along 18°N with MISR overpass
B30329 JuneMainly high-altitude radiation measurements

2. Satellite retrievals of dust aerosol

Most satellite sensors include measurements in the visible portion of the electromagnetic spectrum. Previous-generation instruments such as the advanced very-high-resolution radiometer (AVHRR) have typically provided AOD retrievals in the 600–700 nm range. If clouds can be removed from satellite imagery, then for ocean backgrounds where the contrast between the surface and aerosols are high aerosols can be detected and AOD can be calculated. Multi-spectral measurements are especially well suited to separate aerosols from clouds and the surface. Since the satellite only measures radiances, inverting these measurements to geophysical quantities such as AOD requires a radiative transfer model with assumed surface, aerosol and atmospheric properties. Therefore, for a given sun satellite viewing geometry, AOD can be retrieved from this look-up table approach. AODs retrieved using such methods are usually compared against ground AERONET and aircraft measurements to provide accuracy estimates.

Since the launch of the Terra, Aqua and A-train satellites, significant strides have been made in global aerosol retrievals that have led to the improvement of the study of aerosol effects on climate (Forster et al., 2007). Space-borne sensors dedicated to aerosol science such as the multi-angle imaging spectroradiometer (MISR) (Kahn et al., 2005) and the moderate-resolution imaging spectroradiometer (MODIS) (Remer et al., 2005) have provided unique insights from a multi-sensor perspective (Christopher et al., 2009). Typically satellite remote sensing methods have not provided aerosol measurements over bright targets such as the Sahara Desert, since the visible surface reflectance (Figure 1) is too high to make meaningful interpretations of aerosol signals. Whereas multi-spectral measurements from the visible to the near infrared have excelled in aerosol retrievals over dark targets (ocean and densely vegetated surface), multi-angle measurements from MISR has provided excellent AOD information over deserts (Kahn et al., 2010). However, while MODIS provides near daily coverage due to its large swath (2400 km), MISR provides only weekly global coverage near the Equator because its swath width is about six times smaller.

The total ozone mapping spectrometer (TOMS), which was originally designed to estimate ozone concentrations, is also being used to study UV-absorbing aerosols from space (Torres et al., 2007). Since the surface reflectance in the UV is dark over deserts, TOMS uses spectral information to determine the UV-absorbing aerosol index (AI), which is primarily sensitive to elevated dust and smoke. A new generation of the TOMS instrument called the ozone-monitoring instrument (OMI) uses many of the same principles as that of TOMS but has superior spatial and spectral resolutions. There are two OMI aerosol products available, because two separate algorithms are used: one that primarily uses a two-channel (360 and 380 nm) retrieval algorithm (OMAERO) similar to TOMS (Torres et al., 2007); and the other that utilizes multi-spectral OMI (OMIAERUV) channels (Curier et al., 2008).

Since MODIS has some UV channels, the same principles derived from TOMS and OMI have been used to derive aerosol properties over desert regions, thereby complementing the existing MODIS standard products over ocean and dark targets. This product, called ‘Deep Blue’ (DB), now fills in the gaps in the MODIS standard Collection 5 retrievals (Hsu et al., 2006). PARASOL also provides aerosol products over deserts, although only fine-mode retrievals are available (Herman et al., 2005). The spinning enhanced visible and infrared imager (SEVIRI) on the new generation of METEOSAT satellites also continues to provide qualitative products over the desert regions and some studies are beginning to quantitatively address aerosol retrievals over land from this geostationary satellite (e.g. Brindley and Russell, 2009).

Table I provides a summary of the various satellite products that are now available over bright targets with appropriate references. Note that this is not an exhaustive list of all satellite products over bright targets but only those relevant to this study.

3. Methods and results

In this study, the Level 2 aerosol products from several satellites were collocated in space and time to perform inter-comparisons with AERONET data. AOD values from different satellites was also averaged in a 0.5 × 0.5 degree grid resolution to facilitate inter-satellite comparisons. We provide an overview using the 25 June 2007 case and then summarize the results for the entire GERBILS campaign (18–29 June 2007). The GERBILS flight campaign is fully described in Haywood et al. (2011). Table II lists the various flights during GERBILS, with flight numbers that are referenced throughout the paper.

Figure 1 shows eight flight tracks (Table II) for the GERBILS campaign superimposed on images showing the white-sky broadband (0.3–5 µm) albedos for June 2007 from the MODIS surface albedo product MOD43C3 (Schaaf et al., 2002). These albedos are merely shown to indicate the high surface reflectivity of these regions. Also shown are the locations where AODs were derived by integrating the aerosol extinction coefficient over height during vertical aircraft profiles (denoted by squares with profile numbers) that is used in the analysis. Note that B294 was largely over the Atlantic Ocean and therefore has limited use for this study but can prove useful in radiative closure studies and validation of models developed to simulate dust production, transport, deposition and radiative impacts (Haywood et al., 2011). The B297 flight was also primarily over the ocean and B298 was an evening flight; therefore these flights are not included in the analysis. The flights of particular interest are (B295–B302) from Banizoumbou to Noukachott. The surface albedos are extremely high in some of the regions intersected by these flight tracks. Areas in green generally have surface albedos below 15%, and north of 15°N most areas have high surface albedos (>20%). Most flights travelled along a line of 18°N over Mali and Mauritania, where surface albedos were very high (>0.4). Flight B296 (P2 and P3–P8) took a more southerly track over lower surface albedo regions. Conventional algorithms, such as the MODIS global operational one that relies on dark targets, have difficulty retrieving aerosol properties over this region since the aerosols have reflectivities that are lower than or similar to the background. However, in the UV part of the solar spectrum these surfaces are darker, allowing retrieval of aerosol properties, i.e. DB (Hsu et al., 2006).

Focusing more specifically on conditions surrounding flight B300, Figure 2 shows a true color composite from the MODIS Aqua overpass on 25 June 2007 with red (670 nm), green (550 nm) and blue (470 nm). It is difficult to see dust in this true color composite. Selected AERONET sites are shown as black circles and the flight track for B300 on that day is also shown as a white solid line. Flight B300 on 25 June 2007 started in Nouakchott around 9:25:am and ended around 2:20:pm in Niamey. South of about 12°N, the image has considerable cloud cover, which is shown in white. The inset in Figure 2 shows the AERONET AOD at 550 nm, with high values over Agoufou and Dakar. Missing data due to orbital gaps are shown as a grey area running north to south in this image. The image also shows a thick layer of dust around 20°N and 8°W which produced high AOD values (Figure 3). There was very little cloud cover on the flight track during B300 from Nouakchott to Niamey and the closest AERONET location that B300 crossed was near Agoufou.

Figure 2.

Aqua-MODIS color composite, with red in 0.67 µm, green in 0.55 µm, and blue in 0.47 µm for 25 June 2007. AERONET locations and flight tracks, AERONET mid-visible AODs for selected locations are also shown.

Figure 3.

Mid-visible AOD retrieved from (a) MISR, (b) Deep Blue, (c) estimated AOD from the MISR AOD-AI relationship, and (d) OMIAERO product for 25 June 2007.

Figure 3 shows the spatial distribution of the retrieved AOD from various sensors. Much like AERONET is the gold standard of AOD from ground measurements—MISR—which, because of its expected higher accuracy and relatively lower sensitivity to surface reflectance, is ‘treated’ as a benchmark in this comparison. The multiple look angles coupled with a robust retrieval algorithm (Kahn et al., 2005) have provided excellent comparisons with the AERONET over bright targets such as deserts (e.g. Christopher et al., 2008, and references therein). However, one of the major disadvantages of MISR is its narrow swath width of 360 km compared to MODIS, which is about 2400 km. The flight tracks are also shown in Figure 3(a) as a reference. Figure 3(b) shows the MODIS Deep Blue retrievals and Figure 3(c) and (d) shows the OMI retrievals (Torres et al., 2007) and the estimated AOD (EAOD) retrievals that are derived from relationships between MISR and OMI AI (Christopher et al., 2008; Yang et al., 2009) respectively. Some of the spatial patterns between Deep Blue and OMI are similar. For example, the dust plume with high AODs around Mauritiana and Niger are similar, whereas the OMI product is reporting higher AODs around the coast near Nouakchott that is not seen in Deep Blue. The EAODs in Figure 3(d) are estimated from spatial–temporal regressions between MISR AOD and OMI AI. Therefore the spatial patterns should be similar to Figure 3(c). The EAODs are smoother since they are calculated for every 0.5 × 0.5 degree bins. Also note that in Figure 3(d) AOD values are reported south of 10°N as they were estimated using AI values, whereas Figure 3(c) does not show these retrievals due to cloud masking over that region.

Figure 4 shows the time series of AOD retrievals from SEVIRI for 25 June 2007. Hourly 0.5 × 0.5 degree gridded 550 nm values are shown from 0800 UTC to 1600 UTC. The retrievals are inferred from cloud-free SEVIRI mid-infrared brightness temperatures (Brindley and Russell, 2009) and are performed at a spatial resolution of 3 × 3 SEVIRI pixels, corresponding to approximately 10 × 10 km at nadir. Gridded values are derived if retrievals have been obtained over at least 99% of a given grid box. Clearly, geostationary platforms provide the diurnal information that is not possible from polar orbiters such as Terra or Aqua. Mid-visible 550 nm dust AODs are extremely high throughout the day but there are some interesting features around the flight path. The spatial distribution of the dust plume in southern Mali and Mauritania at 0800 is weak and continues to build in time, both in spatial extent and in AOD up to 1400 UTC. The spatial distribution and concentrations become weaker around 1600 UTC. This diurnal nature of aerosols can only be captured by ground measurements from AERONET (which are limited) and by geostationary satellites. These datasets have been used to assess diurnal aerosol effects (Brindley and Russell, 2009). Another advantage of using geostationary satellites for dust aerosol is that the sequence of images can reveal where and when dust plumes first begin, thereby allowing the identification of dust source regions (Schepanski et al., 2009).

Figure 4.

Mid-visible AOD retrieved from SEVIRI for 25 June 2007 for 0800–1600 UTC.

To get a sense of how the satellite retrievals compare with AERONET AOD, Figure 5(a)–(d) shows AERONET comparisons with various satellite products (MODIS, EAOD, OMEARUV, and SEVIRI). This analysis is conducted using three AERONET sites for which version 2.0 data were available at the time of analysis. These locations include Dakar, Banizoumbou, and Ilorin. Also note that these comparisons were performed only for the GERBILS time period (18–29 June 2007). A more extensive comparison can be seen in validation papers by various science teams (see references in Table I). The AERONET values within ±30 min of each satellite overpass time were averaged, whereas satellite data were obtained by averaging all valid AOD retrievals within a ±0.25° grid box centered on the AERONET site. All satellite algorithms perform well when compared to AERONET, although one could question that the number of data points are small for this comparison. The correlation coefficients between AERONET and MODIS Deep Blue, EAOD, OMEARUV, and SEVIRI are 0.76, 0.88, 0.91, and 0.92 respectively, and slope values are close to one for all sensors. Very few MISR points were available over these three sites for the study period and therefore correlation coefficients are not reported, although MISR AODs are shown as green dots in Figure 5(b). These results are encouraging considering that the surface conditions are extremely bright (Yang et al., 2009), and the contrast between the surface and aerosols are not optimal for most of the satellite retrievals.

Figure 5.

Scatter-plot of mid-visible AERONET AOD versus (a) Deep Blue, (b) EAOD (also shown are the MISR points), (c) OMIAERO, and (d) SEVIRI for 18–29 June 2007.

Figure 6(a) and (b) shows a comparison between the aircraft-derived AODs and EAOD and Deep Blue AOD respectively. Different colors in the scatter-plot represent different flights during GERBILS field experiments. A similar comparison was performed for the dust and biomass-burning experiment (DABEX) (Johnson et al., 2009). The FAAM AOD was estimated by integrating the measured aerosol scattering profile over height. Adjustments for aerosol absorption were also made based on column single scattering albedo from the PSAP data. Johnson et al. (2009) also note that when the aircraft is making a vertical profile, it covers a horizontal path of about 100 km that is several satellite pixels wide and therefore is only an approximation to a true column measurement. Although linear correlation coefficients are high, the satellite retrievals report smaller AODs at larger optical depth ranges, which is clear from lower slope values of 0.87 and 0.76 for EAOD and DB AOD respectively. For AOD < 1, there is good agreement between the satellite retrievals and the in situ values, but in general satellite retrievals underestimate AODs at larger AOD values (Table III). Satellite algorithms have difficulty detecting thick dust plumes and are sometimes classified as clouds, thereby avoiding AOD retrievals. This could be a reason for the differences (Christopher et al., 2009).

Figure 6.

Comparisons of aircraft AOD versus (a) OMI EAOD and (b) Deep Blue AOD.

Table III. Mid-visible aerosol optical depth from aircraft and various flight products.
Flight #ProfileLat.Long.Alb.Flight AODMISRDBAODEAODOMISEVIRI
  1. Alb., broadband MODIS white sky broadband albedo; DBAOD, Deep Blue aerosol optical depth; EAOD, estimated AOD from MISR– OMI relationship.

b294P420.5−17.70.0320.457  0.786  
b294P517.8−16.10.2380.432 0.6130.7270.977 
b295P117.9−15.20.4050.6270.70.5291.0751.387 
b295P416.2−1.60.3381.121 0.7530.9741.5502.196
b295P5–714.80.70.3241.213 1.2480.6091.120 
b295P813.72.10.2860.928 0.6870.7370.656 
b296P114.01.60.3130.294 0.3720.4530.325 
b296P213.0−8.40.2012.236 1.726   
b296P3–813.3−11.80.1781.632  1.0881.8282.594
b296P916.6−16.00.2630.649 0.5650.9240.678 
b299P113.91.60.3130.4640.5790.5880.6060.490 
b299P2–317.5−0.90.4031.342 0.6340.9490.8331.351
b299P4–518.0−2.90.4240.918 0.4830.8450.6980.599
b299P617.9−11.10.4080.554 0.5320.8030.8880.738
b299P918.3−16.10.2760.613 0.804   
b300P118.0−15.20.4050.684 0.5000.8181.0080.834
b300P218.0−7.70.4230.6910.5400.4920.6800.5670.306
b300P3–418.0−6.50.4510.7550.5590.2080.9221.0980.894
b301P113.91.50.3130.593 0.644   
b301P217.4−2.80.3971.6840.7300.7551.0601.4522.292
b301'P417.9−12.30.3880.565 0.4980.6700.7120.685
b301P518.1−15.20.4050.823 0.8560.7411.0671.024
b302P117.9−15.50.4040.5950.6340.7480.6350.5030.542
b302P217.9−13.30.3870.5370.5710.4820.6350.4900.422
b302P317.7−4.40.4320.451 0.5100.430  
b302P614.01.40.3131.099 0.9540.9491.298 
b303P114.01.50.3130.545 0.6060.6250.415 

Figure 7 shows how the AOD retrievals from various satellite platforms perform when compared to the MISR for the entire study region in 0.5 × 0.5 degree grids. Note that the number of data points used for calculating the AODs are different for the various algorithms, dependent on various factors including swath width, cloud screening, and the actual retrieval themselves. Daily half-degree gridded AOD values were obtained from each satellite sensor and presented as a function of surface albedo considering MISR AOD as a benchmark in the region. Comparisons were made for only those pixels (grid box) where all five AOD (MISR, OMI, EOMI, DB, SEVIRI) values were available during GERBILS. The results are separated by the MODIS surface albedo to see if there are systematic biases or differences. The EAOD compares extremely well (r = 0.74) with the MISR AOD (Figure 7(a)), but that is to be expected since it is the monthly mean MISR AOD and OMI AI regression relationships that were used to obtain the AOD for each OMI AI value during the study period. Most of the data points are close to the one-to-one line, except for a few higher values of MISR AODs were underestimated by EAOD. The slope value of 1.11 represents overall overestimation by EAOD. There appear to be no systematic biases as a function of surface albedo in Figure 7(a), except that AOD values over high surface albedo (>0.3) look more scattered than AOD over low surface albedo (<0.3). The Deep Blue AODs (Figure 7(b)) show larger scatter (linear correlation coefficient, r = 0.3) when compared to the MISR. We checked the various cloud flags in the DB product and this result does not appear to be a function of cloud cover. It also appears that for larger surface albedos (0.3–0.4 and 0.4–0.6) the DB values are smaller than the MISR and therefore off the 1:1:line. The slope value of 0.82 indicates an overall underestimation by DB AOD in this region during the GERBILS time period. Figure 7(c) shows that the OMAERUV product also has a higher scatter when compared to the EAOD MISR relationship. The OMAERUV product shows some very high AOD values (1.0–3.0) corresponding to lower MISR AOD values (0.7–1.0) and which could be due to possible cloud contamination in the OMAERUV product (a limitation of OMI due to its large pixel size). The SEVIRI AOD values provide a similar comparison (r = 0.63, slope = 1.22) to OMAERUV but here its dependence on surface albedo is more visible. Under very high surface albedo (>0.4) SEVIRI AOD values are very low compared to MISR AOD values. For a surface albedo range of 0.2–0.4, SEVIRI reports higher AODs than MISR at high AOD values and lower AODs than MISR at low AOD values. In all of these 550 nm satellite retrievals, there are two major sources of uncertainties: surface characterization and selection of the aerosol model for retrieval. In the case of the OMAERUV product the 360 nm and 380 nm wavelengths are used to report AODs at 550 nm that could induce another source of uncertainty. If we assume that the MISR values are the benchmark based on prior studies (e.g. Christopher et al., 2008; Yang et al., 2009) then the AOD differences as a function of surface albedo suggest that there room for improvement in the satellite algorithms investigated here.

Figure 7.

Intercomparison of MISR AOD versus (a) EAOD, (b) Deep Blue AOD, and (c) OMIAERO for 18–29 June 2007 for 0.5 × 0.5 regions for study area between 10–30°N and 20°W–10°E. These relationships are categorized as a function of MODIS white sky albedo ranges shown in different colors.

4. Summary and conclusions

This paper provides a remote sensing perspective of various satellite products that are now available over the North African desert regions. Previous generations of satellite algorithms were not able to deliver AOD products over bright surface reflectance regions such as the Sahara Desert because the contrast between the surface and the aerosol layer was small. New techniques include multi-angle techniques (MISR), use of the UV spectrum (OMI/Deep Blue), infrared geostationary observations (SEVIRI) and estimations based on multiple sensors (MISR-AI EAOD relationship), and these now provide routine retrievals of aerosol properties under cloud-free conditions. This paper evaluates the satellite products mentioned above during the GERBILS period (18–29 June 2007) by comparing products in 0.5 × 0.5 degree grids for the study region of (10–30°N and 20°W–10°E). We conclude the following:

  • 1.All satellite AOD products match well with AERONET retrievals from three locations during GERBILS.
  • 2.Satellite AOD values compare well with aircraft derived AODs for low AOD (AOD < 1) but underestimate at high AOD values.
  • 3.Judged against MISR as a benchmark for AOD retrievals over bright targets, the estimated AOD from the AI-MISR AOD relationship performs best for the entire study region. Both OMAERUV and SEVIRI AOD appear to either overestimate or underestimate AODs at higher surface albedos.

Although MISR AOD retrievals are of excellent quality over deserts, their spatial extent is limited due to the narrow swath of the sensor (360 km). Since atmospheric modeling studies require daily distributions of AOD (not just a limited 360 km region) the estimated AODs from the AOD-AI relationships appear to be a good choice for validating models.

Acknowledgements

This research is supported by NASA's Radiation Sciences, Interdisciplinary Sciences, and ACMAP programs. The MODIS data were obtained through the Goddard Distributed Active Archive Center. We thank the AERONET PIs for providing the data. Special thanks to Christina Hsu for providing the Deep Blue AODs. The FAAM aircraft is jointly funded by the Met Office and the UK Natural Environment Research Council.

Ancillary