This paper aims to evaluate CALIOP aerosol optical depth (AOD) retrieval using MODIS AOD with the goal of improving the CALIOP selection of the lidar ratio leveraging the vertical resolved CALIOP and multispectral MODIS observations. Comparing the MODIS fine mode ratio to CALIOP, we find that the CALIOP integrated attenuated total color ratio provides sensitivity to the aerosol size and type. This finding can be used to better constrain the lidar ratio and improve the CALIOP AOD independent from MODIS.To retrieve the aerosol optical depth from CALIOP requires knowledge of the aerosol lidar ratio that varies significantly as a function of aerosol type. For most CALIOP retrievals the lidar ratio is estimated by correlating CALIOP observables (depolarization and backscatter) with precomputed lidar ratio climatologies. Applying a lidar ratio not representative of the observed aerosols can result in significant AOD biases and is one of the primary sources of uncertainty in the current CALIOP AOD. We demonstrate that over ocean the MODIS sensitivity to the fine- and coarse-mode aerosol mixing ratios provides additional constraints to the aerosol lidar ratio. When MODIS fine-mode retrievals are collocated with CALIOP, the improved lidar ratio significantly reduces the CALIOP AOD mean biases from ∣0.064∣ to ∣0.020∣ when compared to MODIS. In addition, we demonstrate that the CALIOP integrated attenuated total color ratio is correlated with the MODIS fine and coarse mixing ratios in marine environments. This finding suggests that for a CALIOP-only AOD retrieval the integrated attenuated total color ratio can be used to better constrain the lidar ratio. Using the CALIOP integrated attenuated total color ratio, the CALIOP-only AOD retrieval improves the AOD mean biases (∣0.064∣ to ∣0.007∣) when compared to the MODIS AOD.
 Accurate global characterization of aerosol optical depth (AOD) is essential to quantify the energy balance for climate change studies [Charlson et al., 1992]. In terms of climate forcing, aerosols can result in both direct and indirect effects, with the indirect being the most difficult to observe. A commonly studied indirect effect occurs when aerosols change the cloud albedo by modifying cloud drop number concentration [Twomey, 1991]. Observing this effect using passive sensors such as Moderate resolution Imaging Spectroradiometer (MODIS), has proven challenging because of the difficulty in retrieving simultaneous cloud and aerosol properties from passive observations [King et al., 1992].
 The recent success of space-based lidar observations provides an opportunity to advance our understanding of these cloud/aerosol interactions. Active sensors such as Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) aboard CALIPSO (Cloud and Aerosol Lidar and Infrared Pathfinder Spaceborne Observations) can derive both cloud and aerosol properties simultaneously, providing the potential to quantify the aerosol-cloud interactions (indirect effect) [Winker et al., 2009]. However, as will be discussed, accurate retrievals of AOD from CALIOP has proven challenging because of the difficulties in constraining the extinction to backscatter ratio (S or lidar ratio) with the current lidar ratio selection method developed by CALIOP described by Omar et al. . Since the uncertainty in the lidar ratio can significantly affect the accuracy of the aerosol extinction retrieval, lidar ratios have been constrained by many approaches. For example, Reagan et al.  demonstrated an approach based on aerosol models to limit uncertainty in assumed aerosol types. Recently, McPherson et al.  evaluated the Constrained Ratio Aerosol Model-fit (CRAM) technique using measurement from AERONET (Aerosol Robotic Network) [Holben et al., 1998] and HSRL (High Spectral Resolution Lidar) data [Hair et al., 2008]. In this approach, CRAM constrains the aerosol types requiring the spectral ratio of retrieved aerosol properties to fall within the range of aerosol properties derived from AERONET. Müller et al.  also compare the lidar ratio statistics derived from Sun photometer (based on modeling from 180 degree scattering) and Raman lidar measurement. The comparison covered major aerosol types such as marine, desert dust, aged biomass burning smoke and urban/industrial and found significant lidar ratio differences in some aerosol types [see Müller et al., 2007].
 Recent comparisons of CALIOP version 2 AOD and MODIS AOD can be found in work by Kittaka et al. . The CALIOP version 3 AOD (see Figures 1a and 1b) show significant improvements when compared to MODIS. Details changes of version 3 algorithm and code can be found in work by Vaughan et al. [2010b]. Recently, Yu et al.  compared CALIPSO aerosol observation with aerosol simulations from the Goddard Chemistry Aerosol Radiation Transport (GOCART) model and found that the CALIPSO lidar ratio generally agrees with GOCART model within 30% for regional scales. For small AOD, the AOD fractional error can be approximately estimated as the fractional uncertainty of the lidar ratio. For larger AOD, the fractional uncertainty increases (i.e., fractional uncertainty of ∼50% for AOD = 0.5 and nearly 100% for AOD = 1.0) [Yu et al., 2010]. In-depth comparisons with other parameters (such as aerosol layer height, aerosol extinction near source region) can be found in work by Yu et al. .
 In this manuscript we present a comparison between the collocated MODIS and CALIOP (version 3) aerosol retrievals. We find a significant bias between the MODIS and CALIOP AOD, with CALIOP underestimating the AOD relative to MODIS with the bias likely resulting from uncertainties in the CALIOP selected lidar ratio. This finding motivates our investigation into improving the lidar ratio selection process. Previous findings [Sasano and Browell, 1989] demonstrate the potential of a multiple wavelength lidar for discriminating between several aerosol types such as maritime, continental, stratospheric, and desert aerosols on the basis of wavelength dependence of the aerosol backscatter coefficient, which is mainly dependent on the aerosol size distribution and refractive index. Combined lidar and MODIS (derived aerosol properties) to derive the fine-/coarse-mode aerosol extinction profile has been demonstrated by Kaufman et al.  and Léon et al. . We build on this research by demonstrating that the MODIS multiwavelength sensitivity to aerosol size can be used to better constrain the lidar ratio, thereby significantly improving the CALIOP AOD retrievals in marine environments. This approach is fundamentally different from that of Burton et al. , which used the MODIS AOD directly to constrain the CALIOP AOD. Comparing the MODIS retrieved fine-mode ratio to the CALIOP integrated attenuated total color ratio, we find the CALIOP color ratio is correlated with the MODIS coarse mixing ratios in marine environments. This finding suggests that for a CALIOP only AOD retrieval, the integrated attenuated total color ratio can be used to better constrain the lidar ratio, and improve the accuracy of the retrieved AOD independent of MODIS.
 This paper is organized as follows. In section 2, the CALIOP AOD retrieval algorithm is briefly discussed. In section 3, we show the direct comparison of collocated MODIS and CALIOP sensor pixels over land and ocean. In section 4 we present the lidar ratio selection using a combined CALIOP/MODIS retrieval approach. In section 5 we assess the combined MODIS-CALIOP AOD retrieval. In section 6, we show the improvement in AOD retrieval using integrated attenuated total color ratios in the modified CALIOP algorithm over ocean and then summarize the findings in section 7.
2. Retrieving Aerosol Optical Depth Using CALIOP
 In this section, we describe the elastic backscatter lidar equation and the relationship between the measured backscattered signal (or derived integrated attenuated backscatter), the lidar ratio, and the aerosol optical depth (AOD). The CALIOP sensor measures the attenuated backscatter signal, β′(r), which is the product of the total (molecule + aerosol) volume backscatter coefficient, β(r) at range r and the (molecule + aerosol) two-way transmission T2(r), between the lidar and the targeted volume backscatter. The volume backscatter coefficient β(r) is the product of the scattering cross-section per unit volume from range r and backscatter phase function from range r (i.e., where 3/(8π) and ρ(π, r) are the molecular and aerosol backscatter phase function (at scattering angle = 180 degree) from range r; σ′m(r) and σ′a(r) are the molecular and aerosol scattering cross-section per unit volume from range r. The two-way transmission can be defined as T2(r) = exp, where σT is the total (molecule + aerosol) extinction coefficient. Both the molecular and aerosol backscatter contribute to the atmospheric extinction and backscatter coefficient as σT(l) = σm(l) + σa(l) = Smβm(l) + Saβa(l), where Sm and Sa are the lidar ratio of molecules and aerosols.
 The signal detected by elastic-backscatter lidar can be described as:
P is the measured signal after subtracting the background and multiple-scattered return at range r; E0 is the average laser energy for the single-shot or composite profile; ξ is the lidar system parameter; the two-way transmission can also be written as T2(r) = exp(−2 τ(r)),where τ is the total optical depth (molecule + aerosol) at range r, which can be defined as
The CALIOP AOD retrieval is not constrained in equation (1) because both the volume backscatter coefficient and two-way transmission or feature optical depth (or extinction coefficient) are unknowns and cannot be directly separated from the measured attenuated backscatter. To retrieve the aerosol optical depth, one must assume a relationship between the aerosol extinction and the volume backscatter coefficient. The lidar ratio defines this relationship but is difficult to estimate precisely because of its strong dependence on the aerosol size (habit), shape and composition (i.e., absorption or complex refractive index of aerosol) [Müller et al., 2007; Cattrall et al., 2005]. Therefore particle size alone cannot distinguish aerosol type but having additional information of aerosol size can improve determination of aerosol type. The defined lidar ratio can differ by more than a factor of two depending on the aerosol type (i.e., marine S = 20 sr, smoke/biomass burning S = 70 sr). As a result, if the CALIOP retrieval applies the wrong aerosol type, significant biases can result in the retrieved AOD.
 Using the lidar ratio from the CALIOP selection algorithm and the measured backscatter signal, the optical depth can be retrieved using equations (3a) and (3b). The CALIOP feature integrated attenuated backscatter can be computed with aerosol optical thickness according to formula derived by Platt et al. :
which can be rewritten as
where γ′feature is the feature integrated attenuated backscatter, τ is the feature AOD, η is the multiple scattering factor, and Sa is the aerosol lidar ratio. As shown in equation (3b), the CALIOP-derived AOD has a nonlinear dependence on the lidar ratio. Since the CALIOP algorithm assigns the lidar ratio from six aerosol types, an incorrect lidar ratio selection can result in significant biases in the retrieved AOD. It is the focus of this manuscript to both characterize the accuracy of the CALIOP AOD retrievals and then investigate methods to improve the selection of the lidar ratio leveraging both the CALIOP and MODIS sensitivity to aerosol particle size.
3. Direct Comparisons of MODIS and CALIOP AOD
 Both MODIS and CALIOP sensors are part of the Earth Observing System (EOS) “A train” constellation (http://atrain.nasa.gov/), allowing for direct comparisons of MODIS and CALIOP observations. However, there are challenges in merging the two observations since (1) there are significant differences in the spatial and temporal sampling (2) CALIOP observing instrument is approximately nadir viewing and MODIS is cross scanning, and (3) the CALIPSO orbit trails MODIS (Aqua) by approximately 80 s.
 We leverage collocation software developed by Nagle and Holz  at the Space Science and Engineering Center (SSEC) at the UW-Madison to retrieve collocated pixels of MODIS and CALIOP. More details of the collation software and the scheme for deriving collocated pixels from MODIS (Aqua) and CALIOP can be found in work by Nagle and Holz  and Holz et al. .
 To evaluate MODIS (Aqua) and CALIOP collocated AOD, we chose MODIS Collection 5 (C005) Aqua Level 2 (MYD04) 10 km resolution (which is ‘Optical_Depth_Land_And_Ocean’ at 0.55 μm) and ‘Feature_Optical_Depth_532’ of CALIOP Level 2 (CAL_LID_05kmAlay-Prov-Version 3) for 123 days of data (January (01-01-2007 to 01-31-2007), April (04-01-2007 to 04-30-2007), July (07-01-2007 to 07-31-2007), and October (10-01-2007 to 10-31-2007). We also filtered the data using the following criteria: (1) Fraction of cloud cover from both MODIS Aqua (MYD06) and CALIOP (CAL_LID_5kmClay-Prov-Version 3) in the collocated pixels must be less than 0.01. (2) Only a single layer aerosol case as determined by CALIOP. (3) Layer Integrated Attenuated Backscatter (IAB) Quality Assurance Factor must be greater than 0.9.
 After filtering, approximately 7.3% of land and 3% of ocean data is retained. The cloud mask and IAB quality assurance filters are responsible for approximately 87% over land and 95% over ocean of the rejected retrievals. Next, the residuals are filtered for single layer aerosols; approximately 45% over land and 40% over ocean cases are rejected using this filter. A description of the MODIS and CALIOP cloud detection algorithm can be found in work by Ackerman et al.  and Vaughan et al. . As presented in Figure 1, the MODIS and CALIOP retrieved AOD do not agree well with the one standard deviation of the percentage differences of MODIS and CALIPSO AOD being approximately 1400% and 95% for land and ocean respectively. The MODIS AOD has been extensively evaluated over land and ocean using comparisons with ground based AERONET stations. Based on this evaluation the expected uncertainty of the MODIS-derived AOD compared to ground truth (AERONET) is well documented as Δτ = ±0.03 ± 0.05 τMODIS over ocean and Δτ = ±0.05 ± 0.15 τMODIS over land [Remer et al., 2008; Levy et al., 2009, 2010]. Moreover, MODIS data has been evaluated with numerous airborne field campaigns (for example, Chesapeake Lighthouse and Aircraft Measurements for Satellites (CLAMS) in 2001 [Levy et al., 2005], Megacity Initiative Local and Global Research Observation (MILAGRO) in 2006 [Redemann et al., 2009]). Based on this heritage, we use the MODIS AOD to investigate the CALIOP retrievals.
 We find that, in general, the CALIOP-derived AOD is lower than that from MODIS over both land and ocean as presented in Figure 1. To investigate, we focus on ocean retrievals where MODIS is best constrained. Figure 2a presents the histogram of the lidar ratio used in the CALIOP AOD for ocean cases. The CALIOP defined lidar ratio (Sa) at 532 nm for aerosol models are as follows: marine = 20 sr, background = 35 sr, desert dust = 40 sr, polluted dust = 55 sr, smoke or biomass burning = 70 sr, polluted continental = 70 sr. We will further investigate the CALIOP lidar ratio selection method in section 4.
 We find that the marine aerosol model (where lidar ratio = 20 sr) is one of the most common aerosol types over ocean. The direct comparison between MODIS and CALIOP-derived AOD for the marine aerosol cases is presented in Figure 2b. The CALIOP-derived AOD is significantly lower than the MODIS-derived AOD with τCALIOP = 0.24 * τMODIS + 0.02, and a correlation coefficient of R = 0.59. As previously discussed, the CALIOP-derived AOD is strongly dependent on the assumed lidar ratio suggesting that the bias found in Figure 2b results from CALIOP applying a lidar ratio that is too small. At 532 nm the lidar ratio is strongly dependent on aerosol particle size [Müller et al., 2007; Cattrall et al., 2005]. As will be discussed, the multispectral MODIS observations have sensitivity to the aerosol fine- and coarse-mode ratio, with the potential to better constrain the lidar ratio, and CALIOP AOD retrievals.
4. Investigating the CALIOP Selection of the Lidar Ratio in Marine Environments Using MODIS
 In the CALIOP algorithm, the lidar ratio is selected from a look-up table (LUT) that depends on the observed characteristics of the aerosol layer such as the layer altitude, location, volume depolarization ratio δ, the layer integrated attenuated backscatter γ′, together with the surface type. The LUT has been generated from AERONET [Holben et al., 1998] observations using a cluster analysis [Omar et al., 2005, 2009]. There are six aerosol lidar ratios in the LUT, which are marine (Sa = 20 sr), background (Sa = 35), desert dust (Sa = 40 sr), polluted dust (Sa = 55 sr), biomass burning (Sa = 70 sr), and polluted continental (Sa = 70 sr). For the limited cases when aerosols are in a lofted layer with clear air below, Sa is computed directly from the integrated backscatter and transmittance. More details of the lidar ratio selection algorithm are provided by Omar et al. .
 The challenge for CALIOP is that the current Sa selection algorithm does not use criteria that are directly related to the aerosol size which is directly related to the optimal lidar ratio. In this section, we investigate the relationship between the AOD bias, lidar ratio, and particle size using collocated MODIS and CALIOP observations.
Figure 3a presents the normalized occurrence of the different aerosol types used in the CALIOP algorithm as a function of the MODIS AOD fine-mode ratio. The MODIS fine-mode ratio or FMF (fine-mode fraction) is defined in equation (4), with a detailed description available from Levy et al. [2009, 2010]. In Figure 3b the biases between the MODIS and CALIOP-derived AOD are presented as a function of the CALIOP Sa selection.
 In Figure 3a, we expect that the CALIOP marine aerosol selection to be strongly correlated with small MODIS fine-mode ratios (i.e., large particles) since marine aerosol types are generally coarse. However, we find that when CALIOP selects the marine aerosol lidar ratio MODIS often detected a mixture of fine and coarse aerosols. Incorrectly defining the aerosol type as marine and missing the fractional presence of fine-mode aerosols in the scene results in an underestimation of the lidar ratio and CALIOP underestimating the AOD relative to MODIS as shown in Figures 2b and 3b. Further investigation of this bias and methods to improve the CALIOP retrieval will be presented in the next section.
 We also find disagreement between CALIOP and MODIS for polluted dust cases. The CALIOP defined polluted dust model is a combination of desert dust (coarse mode) and smoke (fine mode) aerosol models [Omar et al., 2009]. We expected to observe that the polluted dust cases are associated with the mixture of fine and coarse aerosol with coarse mode accounting for approximately 2/3 of the distribution [Omar et al., 2009, Figure 1]. This is not the case for the fine aerosol distribution that is high as a function of MODIS fine-mode ratio (Figure 3a). However, we find that the cases with large (AOD) biases between MODIS and CALIOP for polluted dust are associated with the MODIS fine-mode ratio as shown in Figure 4b and Table 1.
Table 1. MODIS CALIPSO AOD Differences Versus MODIS AOD Fine-Mode Ratio or FMF for CALIOP Polluted Dust Cases
MODIS-CALIPSO AOD Differences (X)
FMF ≤ 0.5 (460 Cases)
0.5 < FMF < 0.7 (898 Cases)
FMF ≥ 0.7 (1218 Cases)
X ≤ −0.1
−0.1 < X < + 0.1
+0.1 ≤ X
 The correlation between MODIS and CALIOP AOD for the polluted dust is presented in Figure 4a. Notice there is scatter between the MODIS and CALIOP-derived AOD. We find that the biases are associated with 3 different thresholds of MODIS AOD small ratios as presented in Figure 4b. For cases when the MODIS fine-mode ratio or FMF is small (fine-mode ratios ≤ 0.5) the CALIOP AOD is often overestimated. This result is consistent with assuming an aerosol size that is too small (overestimating the lidar ratio). On the other hand, for MODIS AOD fine-mode ratios ≥ 0.7 (the fine-mode aerosols dominates), CALIOP AOD is underestimated due to underestimating the Sa with the Sa ratio of smoke (Sa = 70 sr) being more appropriate for these cases. The mean bias, standard deviation, and details of Figure 4b (i.e., polluted dust cases) are presented in Table 1. The next section presents a combined MODIS/CALIOP retrieval that leverages the MODIS particle size sensitivity to better constrain the CALIOP lidar ratio selection.
5. A Combined MODIS/CALIOP AOD Retrieval Algorithm
 As highlighted in section 4, the primary source of uncertainty in the CALIOP AOD retrieval is the difficulty constraining the lidar ratio that is strongly correlated with the aerosol size distribution. In this section we leverage the MODIS sensitivity to the fine and coarse-mode ratio to constrain the CALIOP lidar ratio in marine aerosol cases over ocean.
 In the current retrieval the lidar ratio is determined using just the CALIOP observations, details can be found in work by Young and Vaughan . The retrieval uses an algorithm and a statistical LUT to estimate this ratio [Omar et al., 2009] and as presented in section 4, significant biases can result from this approach. Using collocated CALIOP and MODIS observations and leveraging the MODIS sensitivity to particle size offers the potential to improve the CALIOP AOD retrievals as the lidar ratio is strongly correlated with the particle size.
 This combined MODIS-CALIOP AOD retrieval separates the optical depth as a function of the fine- and coarse-mode ratio retrieved by MODIS as presented in equation (6). The lidar ratio is selected from the CALIOP LUT but with the ratio determined by the MODIS fine-mode ratio. For small particle sizes, there are two possible lidar ratio values in the CALIOP LUT (70 and 55 sr). For large particle sizes a lidar ratio 20 sr is selected. The feature integrated attenuated backscatter γ′ is obtained from the CALIOP Level 2 profile product. The multiple scattering is assumed to be one to be consistent with the algorithm.
 The combined CALIOP/MODIS retrieval was run for 123 days of data (Jan, Apr, Jul and Oct 2007) consistent with the comparisons presented in Section 3. To investigate the two possible lidar ratios for the three small-mode aerosol types (i.e., Both smoke and polluted continental S = 70; and polluted dust S = 55), the retrieval was processed using both lidar ratios (70 and 55 sr) with the results compared to the MODIS AOD in Figure 5.
 The correlation between the AOD from the MODIS-CALIOP combined retrieval and the MODIS AOD is presented in Figures 5a and 5b. The combined retrieval significantly improves the biases in both cases (i.e., the fine-mode lidar ratio assumption 70 sr (Figure 5a) and 55 sr (Figure 5b)) compared to biases between the MODIS and CALIOP retrieval. As shown in Figure (5a)τCALIOP = 0.66 * τMODIS + 0.053, R = 0.49, and Figure (5b)τCALIOP = 0.5 * τMODIS + 0.05, R = 0.42, but the correlation coefficient is lower than the CALIOP algorithm result (i.e., R = 0.59). In addition, the mean bias of the CALIOP algorithm is improved from ∣0.064∣ to ∣0.021∣ and ∣0.019∣ in the combined retrieval algorithm for two lidar ratios (S = 70 and 55 sr) respectively. Moreover, greater than 63% of the data from the combined retrieval algorithm fall within Δτ = ±0.05 and greater then 86.4% of the data fall within Δτ = ±0.1 for two lidar ratios (S = 70 and 55 sr). We also found that the difference between the lidar ratio for the fine-mode 70 sr (smoke and polluted continental) and 55 sr (polluted dust) aerosols have limited impact on the optical properties even though smoke and polluted dust are quite different (absorption coefficient, particle shape, etc.).
 While the combined MODIS/CALIOP retrieval significantly improves the mean bias compared to the CALIOP retrieval, the correlation coefficients are quite small (i.e., R = 0.49 and 0.42). MODIS is a daytime only retrieval allowing only daytime comparisons. Previously the uncertainty in the MODIS fine-mode ratio has been investigated by Kleidman et al.  and Anderson et al. . We also simulated the MODIS FMF uncertainty on the CALIOP AOD retrieval using equation (6) assuming γfeature = 0.005, Sfine = 20, Scoarse = 55, η = 1 with RfineAOD varying between 0.1 to 0.7. This simulates an uncertainty up to 30%. The results are presented in Figure 5c and have been found to be considerably smaller than the scatter found in Figures 5a and 5b (for example, if the uncertainty of RfineAOD is 30%, the percentage differences of derived AOD is less than 20%). It is likely that the relatively weak correlation in Figures 5a and 5b results from the increased solar background which decreases the CALIOP signal to noise (SNR) for daytime observations.
6. Leveraging the CALIOP Integrated Attenuated Total Color Ratio to Improve the CALIOP AOD Uncertainties
 As demonstrated in section 5 the lidar ratio for aerosols is correlated with the MODIS-derived small-mode ratio. This relationship results from the sensitivity of the lidar ratio to aerosol size. Combining the MODIS spectral sensitivity to the small- (fine) and coarse-mode distribution with CALIOP improves the AOD retrievals. The CALIOP feature integrated attenuated total color ratio (equation (7)) also has sensitivity to aerosol particle size providing the potential to improve the CALIOP AOD retrieval independent of MODIS. Selecting the optimal lidar ratio using the CALIOP integrated attenuated color ratio is investigated in this section.
In equation (7), the B1064 and B532 are the attenuated total backscatter coefficient for the 1064 and 532 nm channels with an additional correction for the effect of the molecular and ozone attenuation [Liu et al., 2009]. The CALIOP retrievals currently use χ′ as part of the scene classification algorithm to separate between cloud or aerosol layers [Vaughan et al., 2005; Liu et al., 2004]. For aerosols with average particle sizes less than 5 μm, the backscattering cross section is sensitive to the differences in wavelength between the B1064 and B532 channels as these wavelengths are close to the size of the scattering medium.
 To investigate the sensitivity of the integrated attenuated total color ratio to aerosol particle size, we compare the MODIS fine-mode AOD ratio to the CALIOP integrated attenuated total color ratio in Figure 6. Since both molecular and aerosol scattering contribute to the integrated attenuated total color ratio, only observations when the MODIS AOD is greater than 0.2 are included in the comparison. For these cases aerosol scattering will be significantly larger than the molecular scattering reducing any bias from the molecular contribution when calculating the linear fit between the MODIS fine-mode fraction and the CALIOP integrated attenuated total color ratio in Figure 6.
 In Figure 6 we find a weak correlation between CALIOP and MODIS (correlation coefficient ∣R∣ = 0.54). The integrated attenuated total color ratio is sensitive to both SNR and uncertainties in the 1064 channel calibration [Vaughan et al., 2004, 2005, 2010a]. As previously discussed, during the daytime the CALIOP SNR is significantly decreased due to the increase in solar background. Uncertainties in the 1064 calibration are currently not well understood but would be expected to manifest as a bias in the results. Although we do not expect a perfect correlation, we do expect that the color ratio should be sensitive to aerosol size, offering the potential to better estimate the aerosol lidar ratio. The uncertainties in CALIOP 1064 nm calibration directly impact on the accuracy of CALIOP integrated attenuated total color ratio (see equation (7)) which intern effect both aerosol fine-mode fraction assumption and the CALIOP only AOD retrieval. The uncertainties of AOD retrieval due to the uncertainties of CALIOP integrated attenuated total color ratio are shown at the end of this section.
 Leveraging the relationship between the MODIS fine-mode AOD ratio and the integrated attenuated total color ratio χ′ found in Figure 6, we assume a relationship between the CALIOP integrated attenuated total color ratio and the MODIS fine-mode fraction as χ′ = −0.4 Rfine AOD + 0.76, which can be rewritten as Rfine AOD = (0.76 − χ′)/(0.4). As presented in Figure 7a, we recalculate equation (6) with the assumption that Rfine AOD = (0.76 − χ′)/(0.4) and the lidar ratios for small aerosol and large aerosol are assigned as 55 and 20 sr, respectively.
 We find a weak correlation but a significantly improved mean bias as presented in Figures 7a and 7b. This result is similar to the combined MODIS/CALIOP retrieval presented in Figures 5a and 5b.
 As presented in Figure 7a, the correlation between the new CALIOP AOD using χ′ in the modified retrieval and the MODIS AOD over ocean for marine aerosol model case is τCALIOP = 0.4 * τMODIS + 0.07 and R = 0.34. The result significantly improves the mean bias relative to the retrieval from ∣0.064∣ to ∣0.007∣ but results in a lower correlation coefficient (R = 0.59 versus 0.34). In addition, approximately ∼64.9% of the data from this new method fall within Δτ = ±0.05 and ∼86.2% of the data fall within Δτ = ±0.1 (see Figure 7b). The correlation coefficient between the MODIS AOD and recalculated AOD does not improve. The lack of improvement is likely a result of the lower CALIOP SNR ratio during the daytime. We also examined the sensitivity of the CALIOP integrated attenuated total color ratios to CALIOP only AOD retrieval assuming γfeature = 0.005, Sfine = 20, Scoarse = 55, η = 1 with a 30% uncertainty for a range of integrated attenuated total color ratio (0.32 to 0.76). The results are shown in Figure 7c.
 The uncertainty of CALIOP integrated attenuated total color ratio on the CALIOP only AOD retrieval are on the same order of the scatter found in Figure 7a (for example, if the uncertainty of the integrated attenuated total color ratio is 30%, the percentage differences of derived AOD can be as high as 150%). The result demonstrates the accuracy of the CALIOP only AOD retrieval significantly depends on the accuracy of CALIOP integrated attenuated total color ratio. This result highlights the importance of improving the CALIOP 1064 channel calibration to improve the accuracy of the CALIOP aerosol retrievals.
 Direct comparison between the collocated MODIS and CALIOP (V3) AOD reveals significant biases with CALIOP underestimating the AOD relative to MODIS. Using the MODIS sensitivity to the coarse and fine aerosol distribution, we find that the CALIOP lidar ratio used in the CALIOP retrieval in marine environments is often underestimated. Furthermore, we find that the CALIOP-derived AOD bias is strongly dependent on the aerosol particle size retrieved by MODIS. This finding motivates the development of a combined MODIS/CALIOP retrieval algorithm that uses the MODIS fine-mode ratio to select the CALIOP lidar ratio.
 The combined MODIS CALIOP retrieval significantly improves the CALIOP AOD with respect to MODIS, reducing the mean bias from ∣0.064∣ to ∣0.02∣. In addition, more than 86.4% of the AODs from the combined retrieval fall within Δτ = ±0.1 of the MODIS retrieval. Most importantly, we find a linear correlation between CALIOP integrated attenuated total color ratios and MODIS fine-mode fraction. Leveraging this relationship, a modified CALIOP only retrieval is presented using the CALIOP integrated attenuated total color ratio to select the optimal lidar ratio for marine aerosols. The modified CALIOP only retrieval improves the bias relative to MODIS with the mean bias improving from ∣0.064∣ to ∣0.007∣ with approximately 86.2% of the new CALIOP AODs falling within Δτ = ±0.1 when compared to MODIS AOD.
 The combined CALIOP/MODIS retrieval significantly improves the CALIOP AOD bias relative to the V3 CALIOP AOD products providing aerosol extinction profiles with small biases relative to the MODIS AOD. This retrieval requires collocated MODIS and CALIOP measurements and is limited to daytime ocean observations. The use of the integrated attenuated total color ratio to constrain the CALIOP lidar ratio provides an improved CALIOP AOD retrieval with similar biases relative to the combined CALIOP/MODIS retrieval and does not require the added complexity of collocating the MODIS and CALIOP data. In the current form, it is still limited to ocean only retrievals. In future work, we will investigate developing an integrated attenuated total color ratio relationship for land aerosols using AERONET as a reference data set providing a consistent AOD retrieval for both land and ocean observations.
 We would like to acknowledge the CALIPSO and MODIS projects for providing the CALIOP and MODIS observations and retrievals. We would also like to thank Fred Nagle for his help with the collocation software. This research was supported by NASA grant NNX07AR95G and the Integrated Programs Office (IPO) grant NA06NES4400002.