Estimation of GEOS-Chem and GOCART Simulated Aerosol Profiles Using CALIPSO Observations over the Contiguous United States

Model-simulated aerosol profiles can significantly improve a satellite’s capability to estimate ground-level particle concentrations, but are difficult to validate due to the sparse network of ground-based lidars. We quantitatively evaluated aerosol vertical profiles simulated by the Global 3-D Atmospheric Chemical Transport model (GEOS-Chem) and the Goddard Chemistry Aerosol Radiation and Transport model (GOCART) over the contiguous United States using the CloudAerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO). Model-simulated and satellite-retrieved Aerosol Optical Depth (AOD) are first validated with the Aerosol Robotic Network (AERONET) data. The large discrepancies between satellite and model are due to underestimation of GEOS-Chem AOD in the West and GOCART AOD in the East, along with overestimation of CALIPSO AOD during winter and in the West. Model-simulated Aerosol Extinction Coefficients (AEC) at three layers are evaluated against CALIPSO. Aggregating the data from daily and 2 × 2.5 degree model resolutions to a national or annual scale can significantly improve the correlation coefficient and regression slope. Both GEOS-Chem and GOCART underestimate AEC in the lower troposphere in the East, and in the free troposphere in the West than CALIPSO. We attribute these differences to the coarse horizontal resolution, missing aerosol components, and inappropriate emission inventory of the models. Additionally, the low signal-to-noise ratio and cloud and bright surface reflectance interfere with the satellite’s measurements.


INTRODUCTION
Atmospheric aerosols affect global climate change directly by absorbing and reflecting radiation budget (Myhre, 2009), and indirectly by altering cloud microphysics and biogeochemical cycles (Mahowald, 2011).Satellite remote sensing significantly improves our understanding of aerosol horizontal distribution, but many sensors lack the ability to describe the aerosol vertical structure, a detail that is crucial to link satellite-retrieved column AOD with ground-level PM 2.5 levels.In 2006, the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite was launched as part of the sun-synchronous A-train constellation of satellites (Winker et al., 2007).The Cloud Aerosol LIdar with Orthogonal Polarization (CALIOP) aboard the CALIPSO platform can provide particle extinction and backscatter profiles, as well as particle depolarization ratios.Among the strengths of CALIOP are its high vertical resolution and unambiguous multi-layer detection.However, some deficiencies, such as the nadir-only measurement geometry and low signal-to-noise ratios near ground, limit its large-scale application (Kacenelenbogen et al., 2011).
Atmospheric chemical transport models (CTM) can also expand the sparse ground monitoring networks into regions currently not covered.CTMs have the ability to simulate particle concentrations, speciation, and vertical distribution, in addition to optical properties.A few studies have demonstrated that CTM-simulated aerosol vertical profiles can be used together with satellite data to analyze air pollution transport.For example, Generoso et al. (2008) use the vertically resolved attenuated backscatter from a global 3-D atmospheric Chemical transport model (GEOS-Chem) and MODIS AOD to characterize the Saharan dust outflow over the Atlantic Ocean.The aerosol vertical distribution is also a useful link between satellite-retrieved AOD and ground-level PM 2.5 (Liu et al., 2011), but model-simulated aerosol vertical profiles have not been extensively validated .The launch of CALIPSO provides a new approach to evaluate CTM's aerosol vertical distribution around the world.For example, Yu et al. (2010) grouped the globe into 12 sections that were representative of industrial, dust, and biomass burning pollution regions to examine the Aerosol Extinction Coefficient (AEC) profiles from the Goddard Chemistry Aerosol Radiation Transport (GOCART) model.Johnson et al. (2012) estimated the vertical distribution of dust aerosol extinction from GEOS-Chem by CALIPSO over six dustdominated regions in the world.Ma and Yu (2014) compared the seasonal variability of GEOS-Chem/APM AEC profiles over Eastern U.S. and Western Europe by the CALIPSO level-3 monthly aerosol extinction product.While previous evaluations were often carried out in the source regions or industrial regions at the global or continental scale.To date, almost no studies have specifically focused on the aerosol profiles of GEOS-Chem and GOCART, two wellknown CTMs in a national domain to show their detailed spatiotemporal patterns.In addition, GEOS-Chem and GOCART, combined with satellite retrieved AOD, are widely used in estimating national-scale near surface PM 2.5 concentrations.For example, Liu et al. first developed a scaling method to derive the distribution of annual mean PM 2.5 in the U.S. in 2001 (Liu et al., 2004).Geng et al. (2015) generated a PM 2.5 map at 0.1 × 0.1 degrees over China for 2006-2012 by using the nested-grid GEOS-Chem model, MODIS and MISR data.These studies all treat the aerosol vertical distribution as an important factor to AOD and PM 2.5 correlation.Model simulated aerosol profiles can make up for the inadequacy of ground-based Lidar observation, but their quality also has a large impact on the satellite PM 2.5 estimation.The main objective of this study, therefore, is to evaluate the accuracy of aerosol vertical profiles simulated by GEOS-Chem and GOCART by CALIPSO in the contiguous U.S. under various geographical and climatic conditions.Section 2 describes the datasets and data processing steps.Section 3 gives the summary statistics and the agreement in seasonal and geographical patterns among different data sets.We also discuss the potential sources of uncertainties in detail.Finally, major findings of the current analysis are summarized in section 4.

Data Description
The GEOS-Chem model is driven by assimilated meteorological data from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office.In this analysis, we use the GEOS-Chem version-8.3.2 tracer optical depths at three-hourly temporal resolution, 2° latitude × 2.5° longitude horizontal resolutions and 37 vertical layers from 2007 to 2009.The total AOD of each layer used in this analysis was calculated as the sum of all tracers' optical depth, excluding the coarse mode of dust and sea salt.The GOCART model is driven by the assimilated meteorological fields generated in the GEOS Data Assimilation System (GEOS DAS) (Chin et al., 2000).The retrieval error of GOCART AOD is relatively small in dust and pollution regions, but has a low bias of 30-40% for biomass burning aerosols (Chin et al., 2009).In North/Central America, GOCART AOD is approximately 2 times lower than MODIS (Chin et al., 2004) and MISR (Li, S. et al., 2015).The GOCART 3-D simulations used in this analysis are the same temporal and vertical resolutions as GEOS-Chem, but with higher horizontal resolution (1° latitude × 1.25° longitude).GOCART total AOD at 550 nm is the sum of five tracers including sulfate, dust, organic carbon (OC), black carbon (BC), and sea salt.AOD in both models are calculated using simulated mass concentration of each tracer and their optical properties, which are generated by a Mie code (Martin et al., 2003).The aerosol optical properties in GEOS-Chem and GOCART are based on the Global Aerosol Data Set (GADS), which consists of wavelength-resolved complex refractive indices and estimates of aerosol size distributions at different relative humidity levels (Koepke et al., 1997).
CALIPSO Version-3.01Level-2 cloud and aerosol profile products were collected for the same period and geographic region as the CTMs (accessed at https://www-calipso.larc.nasa.gov/).CALIPSO 5 km product reports both daytime and nighttime profiles of particle extinction and backscatter at 532 nm and 1064 nm over an altitude range of 30 km to -0.5 km, and with a spatial resolution of 60 m (180 m above 20 km) vertically and 5 km horizontally (Young and Vaughan, 2009).CALIPSO retrieval has been accurately accessed by ground-based observation in some case studies (e.g., Kacenelenbogen et al. (2011)), and also has been compared with other satellite data such as MODIS in a large scale (Kittaka et al., 2011).The flags of extinction uncertainty (parameters of 'Extinction Coefficient Uncertainty 532'), quality control ('Extinction QC Flag 532'), feature classification ('Atmospheric Volume Description'), and cloud aerosol discrimination ('CAD Score') are used for data screening.To filter out most of the bad extinction values and keep as many useful samples as possible, we identified a vertical layer as aerosols when the CAD score was less than -20 (personal communication with Jason L. Tackett).
In this analysis, the column AOD from AERONET was used to evaluate the CTMs and CALIPSO.The AERONET is composed of ground-based sun photometers measured in 15-minute time intervals during the daytime.AERONET AOD values are recorded at seven spectral bands (340, 380, 440, 500, 670, 870, and 1020 nm).During our study period, there are 15 AERONET sites in the Eastern U.S. (east of 100°W), and 19 sites in the Western U.S. (west of 100°W) with Level 2 (quality assured) data (accessed at http://aeronet.gsfc.nasa.gov)(Fig. 1).

Data Processing
We processed all input datasets in order to create a coincident dataset with consistent spectral, temporal, vertical, and horizontal resolutions.AERONET AOD at 550 nm is interpolated using 440 nm and 870 nm AOD values using the AERONET-retrieved Angstrom exponent (α 440-870 ).CALIPSO AOD, and AEC at 532 nm is assumed to be comparable with other datasets reported at 550 nm (Kittaka  et al., 2011).Because the mass of the atmosphere mainly resides in the troposphere, the lidar backscatter signals from the stratosphere (> 12 km above the surface) often contain little aerosol information and may have more uncertainties due to the solar background.Following previous comparison studies of Liu et al. (2011) (8 km) and Yu et al. (2010) (10 km), we set an average altitude of 9 km as the upper altitude limit for the two CTMs and CALIPSO.We also divided the troposphere into three layers: the lower troposphere (0-3 km), free troposphere (3-6 km), and upper troposphere (6-9 km).For the comparison of column AOD values, it is necessary to determine an acceptable buffer distance to select CTM and satellites pixels to be matched with each AERONET site.These 2° × 2.5° and 1° × 1.25° grid cells covering the AERONET site are denoted as the "nearest" point for GEOS-Chem and GOCART pixels, respectively.For CALIPSO, Schuster et al. (2012) checked the correlation coefficients between AERONET and CALIPSO as a function of distance, and founded that 80-km horizontal averaging produces result similar to that considering only 5 or 20 km horizontal averages.Thus, this study also chose an 80 km × 80 km box to average the CALIPSO pixels around the AERONET site for comparison.
For the comparison of AEC profiles, we conducted our evaluation at different scales.First, we averaged daily CALIPSO and GOCART pixels in one 2° × 2.5° grid cell (Daily-Grid hereinafter) since GEOS-Chem has the coarsest spatial resolution among all datasets.Second, like the approach of Koffi et al. (2012) that selected 13 subcontinental regions to sample global Lidar aerosol profiles, we used nine National Oceanic and Atmospheric Administration (NOAA)-defined climatically consistent regions in the contiguous U.S. (Fig. 1).These standard regions are identified through analyzing the regional and national monthly, seasonal, and annual temperature and precipitation data during the period of 1895-1983 (http://www.ncdc.noaa.gov/temp-and-precip/us-climate-regions.php).Then, the pixels of Daily-Grid located in each region are processed for regional or national analysis.To correspond to the overpass time (about 1:30 pm) of Aqua and CALIPSO, the threehourly CTM simulations closest to 1:00 pm local time are selected, and AERONET data within a two hour window is averaged around 1:30 pm local time.Seasonal and annual AOD and AEC profiles are calculated from daily data.

Validation of AOD Using AERONET Data
Table 1 shows the summary statistics for the entire AERONET, GEOS-Chem, GOCART, and CALIPSO AOD data, as well as the data stratified by location and season.Overall, there are 12,257 records for 34 AERONET sites Min, Max and Mean are calculated from all matched samples.Slope, intercept and r are fitted between the matched variables (e.g., GEOS-Chem) and AERONET data.over the contiguous U.S. from 2007 to 2009.GEOS-Chem and GOCART provide complete aerosol coverage without any limitations.CALIPSO only captures 652 samples mainly due to the temporal and spatial constraints of its narrow orbital swath and the impact of cloud contamination (Winker et al., 2009).Compared with AERONET (range: 0.0091-0.66;Mean: 0.11) and CALIPSO (range: 0.0054-0.82;Mean: 0.12), CTMs usually show narrower data ranges and lower mean AOD values.Linear regression of AODs from all datasets against the AERONET observations yields the best correlation coefficient (0.63) and slope (0.47) for GEOS-Chem.Geographically, CALIPSO mean AOD are similar to AERONET both in the East (CALIPSO: 0.14; AERONET: 0.14) and in the West (CALIPSO: 0.089; AERONET: 0.082).GEOS-Chem (Mean: 0.054) significantly underestimates AOD (0.082) in the West.GOCART (0.11) underestimates AERONET AOD in the East (0.14).Both CTMs have better correlation coefficients in the East (r_GEOS-Chem: 0.65, r_GOCART: 0.41) than those in the West (r_GEOS-Chem: 0.46, r_GOCART: 0.36).Seasonal mean CALIPSO AOD (0.11) is two-fold higher than that of AERONET (0.056) in winter.On the other hand, the largest discrepancies between CTMs (GEOS-Chem: 0.1; GOCART: 0.11) and AERONET mean AOD (0.15) occur in the summer.Relative to the best correlation coefficients between CALIPSO and AERONET in summer, GEOS-Chem and GOCART show the best agreement in spring (r_GEOS-Chem: 0.7; r_GOCART: 0.56).Bed correlations for all datasets (r_GEOS-Chem: 0.48; r_GOCART: 0.36; r_CALIPSO: 0.2) occurs in winter, which may be due to narrower data ranges and smaller sample sizes, as wells as the impact of bright surface and wintertime atmospheric stability (i.e., limiting the aerosol mixing).

Estimation of AEC at Different Spatial and Temporal Scales
To quantitatively investigate the aerosol vertical distribution of the two CTMs, we calculated the AEC of GEOS-Chem and GOCART in the three tropospheric layers, and compared them with CALIPSO retrievals at different spatial and temporal scales.Overall, there are 32,142 model-satellite matched daily records at 2 × 2.5 degree grid scale from 2007 to 2009 over the contiguous U.S. (Figs.2(a1) and 2(a2)).Daily-Grid scale comparison shows that CALIPSO AEC has a wider value range (0-0.3)compared to the GEOS-Chem (0-0.2) and GOCART estimates (0-0.1).Linear regression of the CTM against CALIPSO AEC yields very low slopes (GEOS-Chem: 0.32; GOCART: 0.19) and poor correlation coefficients (GEOS-Chem: 0.51; GOCART: 0.46).This could be attributed to several reasons.First, CALIPSO cannot provide the continuous profiles as CTMs, especially above 6 km.As shown along the Y-axis of Figs.2(a1) and 2(a2), CALIPSO sometimes shows near-zero values if the layer contains more cloud profiles data.Thus, CALIPSO mean AEC (0.001) in the upper tropospheric layer is lower than GEOS-Chem (0.0013) and GOCART (0.0029).On the contrary, CALIPSO's real-time monitoring capability can capture high AEC when a large amount of aerosols are concentrated at a certain height.It is possible that CALIPSO may have misidentified thin clouds as aerosols in our study, despite the relatively low CAD score we set for screening.However, the aerosol spatial variability is not well simulated by CTMs (Fairlie et al., 2007;Generoso et al., 2008).As seen along the X-axis of Figs.2(a1) and 2(a2), CALIPSO mean AEC (0.008) in the free tropospheric layer is larger than GEOS-Chem (0.0049) and GOCART (0.0067).In addition, although the three datasets show that the AEC in the lower troposphere contribute the largest to the total column AOD, a systematic under-predicting of AEC is still found by the GEOS-Chem and GOCART simulations, with the mean values of 0.033 and 0.024, respectively, compared to 0.041 of CALIPSO.
When comparing the satellite-retrieved and modelsimulated AECs in Daily-Grid scale, the random noise such as from the cloud contamination for CALIPSO or the coarse revolution for GEOS-Chem and GOCART has a large impact on the linear regression result.The slope and the correlation coefficient may be significantly affected by the outliers in Fig. 2(a).Therefore, this approach also aggregates the dailygrid data to national and annual level for a large-scale and long-term comparison.As shown in Fig. 2(b), the number of matched records reduces to 3,039 in Daily-National scale, which also means almost 10 Daily-Grid records are averaged in the contiguous U.S. per day.The correlation coefficients against CALIPSO increase to 0.78 for GEOS-Chem and 0.73 for GOCART, and the slopes increase to 0.59 for GEOS-Chem and 0.37 for GOCART.Fig. 2(c) shows the comparison at the Annual-Grid scale, the number of matched records (501) means that nearly 60 Daily-Grid records are averaged per year.As expected, the temporal averaging improves the correlation coefficient to 0.9 for both GEOS-Chem and GOCART, and the regression slope to 0.72 for GEOS-Chem and 0.44 for GOCART.Our results presented in Fig. 2 indicate the quantitative comparison of model simulations with satellite retrievals should be done with caution because the agreement is usually poor at fine spatial and temporal levels and can be much improved by using coarser spatial and temporal scales.

Seasonal Distribution of AOD and AEC Profiles in Different Climate Regions
Figs. 3 and 4 show seasonal model-simulated and satelliteretrieved results in nine NOAA climate regions, respectively.Fig. 3 indicate that, on average, CTMs have consistent spatial patterns of AOD, with high values persistently located over the more populated and polluted urban regions (e.g., the NOAA Regions 1-3).CTM AOD in the East in summer and spring are significantly lower than CALIPSO retrievals.In addition, CTMs are unable to capture the populated western regions, such as California during all seasons.High AOD values seen in CALIPSO can better match the industrial (e.g., NOAA Regions 1-4) and dust source regions (e.g., NOAA Region 7), but high biases in these retrievals are evident because the correlations between CALISPO and AERONET are bad.
The aerosol concentrations exhibited steep vertical gradients and significant variability in different climate regions (Thornhill et al., 2008;Koffi et al., 2012).The contiguous U.S. includes most climate types.In the East, the climate ranges from temperate continental in the north (Regions 1, 3, and 4 in Fig. 1) to humid subtropical in the south (Regions 2 and 5).In the West, the climate is temperate steppe and alpine in the Regions 6 and 7, desert in the south of Regions 7 and 9, alpine and desert in the inland of Regions 8 and 9, oceanic in coastal Region 8, and Mediterranean in coastal Regions 9. We analyzed the aerosol vertical characteristics by NOAA climate region.
1) In eastern Regions 1-5: in the free and upper troposphere, GEOS-Chem and GOCART are overall comparable with CALIPSO due to the low aerosol loading, with all AEC values less than 0.02.However, the springtime Fig. 4. Profiles of seasonal average aerosol extinction coefficient from GEOS-Chem, GOCART and CALIPSO.The y (elevation) is the altitude above ground level, not sea level.simulated AECs are higher than CALIPSO, especially in Regions 1 and 3.It may be an overestimation of the dust mass in the smallest (0.1-1.0 micron) transport tracer in GC or GOCART.Our findings here are consistent with previous studies of evaluating model dust simulations by observations of ground, aircraft (Fairlie et al., 2007) and satellite (Li et al., 2013).The latest version GEOS-Chem model has considered to change the dust mass partitioning, which can reduce the dust AOD up to 40% downwind where sub-micron aerosols dominate (Ridley et al., 2012).At the lower troposphere, both model and satellite profiles increase rapidly, with all AEC values larger than 0.04.The high percentage of AEC below 3 km is consistent with strong influence from the near surface urban/industrial emissions.The large discrepancies between CTMs and CALIPSO are also found in the lower troposphere.Here, we defined the AEC growth rate as the ratio of lower-troposphere AEC over free-troposphere AEC.For the East and the four seasons, the mean AEC growth rate of CALIPSO, GEOS-Chem and GOCART are 13.78, 9.79 and 4.95, respectively, which indicates CALIPSO AEC generally grows faster than CTMs as they appear to ground.We attribute the discrepancy to the following factors.First and most important, the model underestimation of near surface AEC is caused by the difficulty of obtaining accurate emissions inventory.For example, in high latitude areas (e.g., Regions 1 and 3) in winter, GEOS-Chem (< 0.04) and GOCART (< 0.3) AEC is significantly lower than CALIPSO (> 0.06).Due to the influence of the Labrador Current and cold air from Canada, the anthropogenic emissions from combustion processes in the CTMs may be underestimated during the long winter season.In Region 5, GEOS-Chem (~0.03) is a factor of 2 lower than CALIPSO (~0.06).It might be because the prescribed fires are not well represented by the current Global Fire Emission Database (GFED) used in GEOS-Chem, which is likely because GFED pixel size is too large to capture the small fires.Yet the regional and global anthropogenic emissions inventories need continuous improvement.Since Version 9, the biomass emissions of SO 2 and NH 3 in GEOS-Chem are now accounted for in the GFED inventory.Second, transported aerosols, such as the smoke from Mexico in the spring (Wang et al., 2009), also contribute to the AEC values in the free and lower troposphere, while they may be detected by CALIPSO but not well represented in both models.Third, the tracers such as nitrate and ammonium are missing in the GOCART aerosol composition.It results that GOCART AEC in the lower troposphere in the East is even lower than GEOS-Chem.Fig. 4 also shows the discrepancies between GOCART and GEOS-Chem AEC in Regions 1 and 3 are higher than those in Regions 2 and 5.Both Table 1 and Fig. 3 demonstrate the substantial underestimation seen in the GOCART AOD, which may be explained to a large extent by missing the nitrate and ammonium aerosols.
2) In western Regions 6-9: The profile difference compared to these in eastern Regions 1-5 are mainly from the low AEC values in 3 km, with less than 0.03 for all reasons.The National Land Cover Database (NLCD) (http://www.mrlc.gov/nlcd06_data.php)indicates that the West are mainly covered by scrub, grassland, forest, and barren land, which is less influenced by the fossil fuel usage as compared to populated eastern regions.On the other hand, AEC value of CTM and CALIPSO at the free troposphere is significantly higher than that in eastern U.S. It could be attributed to long-range transport of aerosols from East Asia to the North Pacific Ocean and the North American West Coast in spring (Heald et al., 2006).Yu et al. (2012) found that 140 Tg of dust was exported from East Asia, and 56 Tg of dust reached the West Coast in 2005.In addition to the dust transport, the GFED burn area product shows that concentrated fire activity appear in western states such as Idaho, Oregon, Nevada, and southern California (Morton et al., 2013).Overall, model simulated AECs are lower than CALIPSO in the both free and lower troposphere, but the mean growth ratio of GEOS-Chem (4.05) and GOCART (2.64) are higher than CALIPSO (2.57).There are several possible reasons for this discrepancy.For CALIPSO, it could to a large extent be explained by the errors caused by the bright surfaces, which can increase the noise in daytime signals, contributing to more difficulties in the detection of aerosol layer base height, and resulting in the low bias of aerosol near surface (Kittaka et al., 2011).Over the dust source regions (e.g., NOAA Region 7), the high surface reflectance tends to cause CALIPSO to underestimate AEC.As a result, GC and GOCART simulations appear to be in better agreement with CALIPSO retrievals.Our findings indicate that although CALIPSO can retrieve higher column AOD than CTMs in the West, the percentage of aerosol nearsurface to total column may be underestimated by CALIPSO or overestimated by model.For the model, the aerosol sink assumed in the CTM model may be too rapid due to the strong precipitation in coastal Regions 8 and Region 9 (i.e., as is influenced by the oceanic and Mediterranean climate) (Veefkind et al., 2011).CTMs are also unable to capture AEC in the free troposphere over the dust source area as the NOAA Region 7. It may be that the CTM's dust deposition during plume transport is too strong in the West, as demonstrated by a study of GEOS-Chem dust transport from the Sahara (Generoso et al., 2008).Furthermore, the phase functions of spherical and non-spherical particles are both important for the accurate AOD calculation, especially over dust regions (Wang et al., 2012).But CTM often lacks a satellite's capability to characterize non-spherical particles.The latest GEOS-Chem model has implemented a new wet and dry deposition scheme, as well as the OC growth with relative humidity, which may reduce the uncertainties of aerosol simulation as mentioned above in the western U.S. Despite the high nitrate concentration on the West Coast, primarily in California (Bell et al., 2007), our finding indicates that the discrepancy between GOCART and GEOS-Chem in the West is less than that in the East.It may be because GOCART has a stronger dust or other absorbing aerosol simulation capability than GEOS-Chem (Fairlie et al., 2007), which can make up for the lack of nitrate and ammonium components.
Our work contributes to the current literature from the perspectives of users of various satellite products, which entail a much larger community.By comparing GEOS-Chem, GOCART and CALISPO, we can provide guidance to model developers as well as the CALIPSO science team to improve the accuracy of model and satellite.As mentioned above, the errors in model and satellite data can be attributed to many factors.Our ongoing work will focus on improving their accuracy from the following aspects.First, our validation indicated that both model and CALIPSO AOD are poorly correlated with AERONET observations as compared to those of MODIS and MISR.The ratios of GEOS-Chem/ GOCART/CALIPSO AOD over MODIS/MISR AOD can potentially be used as a scaling factor to change all AEC values in total column.Second, we plan to follow the topdown approach reported in Xu et al. (2013) to spatially constrain aerosol emissions using satellite observed radiances with the adjoint of chemical transport models.Third, model simulated aerosol components as well as satellite-measured surface reflectance may be coupled in the current CALIPSO algorithm to improve the AEC retrieval.

SUMMARY
This analysis evaluated the GEOS-Chem and GOCART column AOD and AEC profiles over the contiguous U.S. from 2007 to 2009 using AERONET and CALIPSO data.We first validated model-simulated and CALIPSO column AOD against AERONET observations.A linear regression analysis demonstrated that GEOS-Chem has the best agreement with AERONET.Second, the CTMs and CALIPSO AEC profiles were compared in the lower, free, and upper troposphere, respectively.Direct comparison of the original AECs yields a low correlation coefficient and slope between CTMs and CALIPSO.The poor agreement is mainly due to the inconsistent profiles of CALIPSO when impacted by cloud, as well as the narrow data range of CTM not captured the instantaneous values in heavily polluted environments.The agreement was significantly improved after aggregating the AEC to the national or annual scale.Third, we compared the GEOS-Chem, GOCART and CALIPSOAEC over nine NOAA Climate Regions and during the four seasons.Overall, all three products have the capability to roughly reflect the anthropogenic and natural aerosol over the contiguous U.S. at a regional and seasonal scale.However, the AEC profiles simulated by CTM are generally lower than CALIPSO retrievals, and we attributed the discrepancy to two aspects.(1) CTMs underestimate local emissions near the surface in the East and most western regions, which can be explained to a large extent by the low bias introduced in the column AOD.(2) CTMs are unable to fully capture the aerosol transport at the free troposphere in the West.Unfortunately, the comparison can only be done at a coarse resolution due to the very sparse coverage of CALIPSO.Using a higher spatial resolution, as we have experimented, resulted in instability in our results.Primary analysis and related literature indicate that these uncertainties are attributed to the errors in emissions inventory, lack of aerosol components for GOCART, coarse spatial resolutions used in CTMs, and the impact of bright surface and low signalto-noise on CALIPSO.Despite the discrepancies among different datasets, a large-scale comparison of satelliteobserved and model-simulated aerosol profiles can still provide useful information on the strengths and weaknesses of both.

Fig. 1 .
Fig. 1.Geographical location of AERONET sites (blue star) in the U.S. standard Climate Regions.Black and red dots are GEOS-Chem 2 × 2.5 and GOCART 1 × 1 degree grid centroids.The CALIPSO path is shown as green lines.

Table 1 .
Summary statistics for AERONET, GEOS-Chem, GOCART and CALIPSO AOD variables by season and geographical regions.Spring is March through May, summer is June through August, fall is September through November, and winter is December through February.