Retrieval of UV-Visible aerosol absorption using AERONET and OMI-MODIS synergy: Spatial and temporal variability across major aerosol environments

Measuring spectral aerosol absorption remains a challenging task in aerosol studies, especially in the 10 UV region, where the ground and airborne measurements are sparse. In this research, we introduce an algorithm that synergizes ground measurements with satellite observations for the derivation of spectral single scattering albedo (SSA, ωo) of aerosols in the UV to visible range (340-670 nm). The approach consists in explaining satellite measured near-UV radiances (340, 354 and 388 nm) by the Ozone Monitoring Instrument (OMI), and visible radiances (466 and 646 nm) by MODerate Imaging Spectrometer (MODIS), in terms of ground-based 15 Aerosol Robotic Network (AERONET) measurements of total column extinction aerosol optical depth (AOD, τ), and retrieved total column wavelength dependent SSA using radiative transfer calculations. Required information on aerosol particle size distribution is taken from an AERONET-based climatology specifically developed for this project. This inversion procedure is applied over 110 AERONET sites distributed worldwide, for which continuous, long-term AERONET measurements are available. Using the derived data set we present seasonal 20 and regional climatology of ωo(λ) for carbonaceous, dust and urban/industrial aerosol types. The UV-Visible spectral dependence of ωo obtained for the three major aerosol types from the synergy algorithm is found to be consistent with the insitu measurements reported in the literature. A comparison to standard AERONET SSA product at 646 nm shows absolute differences within 0.03 (0.05) for 40% (59%) of the compared observations. The derived aerosol ωo(λ) data set provides a valuable addition to the existing aerosol absorption record from 25 AERONET by extending the absorption retrieval capability to the near-UV region. The combined UV-Visible data set, in addition to improving our understanding of spectral aerosol absorption properties, also offers wavelength-dependent dynamic aerosol absorption models for use in the satellite-based aerosol retrieval algorithms. https://doi.org/10.5194/amt-2021-8 Preprint. Discussion started: 19 January 2021 c © Author(s) 2021. CC BY 4.0 License.


30
Atmospheric aerosols play a significant role in the Earth's climate system through scattering and absorption of solar radiation, thus capable of perturbing radiation budget. The ratio of the amount of the light scattering to the total extinction referred to as single scattering albedo (SSA, ω o ) is a fundamental variable used to gauge the absorbing nature of aerosols. Mie-theory indicates ω o equals to one for purely scattering aerosols and less than one towards zero for increasingly absorbing nature of aerosols. Studies show that the estimates of net aerosol 35 radiative forcing is sensitive to the aerosol ω o , and small changes to it could potentially alter the forcing on atmosphere (Chyacutelek and Coakley, 1974;Hansen et al., 1997). Models are often fed with essential aerosol properties to estimate the forcing on the atmosphere. These properties include aerosol optical depth (AOD, τ), complex refractive index, and phase function. Here, the knowledge on spectral dependence of such properties is crucial in quantifying the overall effects of aerosols. For example, absorbing aerosols can lead up to a 50% 40 increase in the near-UV irradiance compared to the similar load of only scattering aerosols in the atmosphere (Bais et al., 2005). A report by Intergovernmental Panel on Climate Change suggests that the lack of spectral aerosol absorption is one of the major contributors leading to significant uncertainties in quantifying the net aerosol radiative effects on the Earth's climate (IPCC, 2013). 45 Developments in ground-based and satellite aerosol retrieval techniques have greatly improved our understanding of atmospheric aerosols over the last two decades. However, knowledge on spectral aerosol absorption properties is limited due to difficulties in measurements (e.g., Heintzenberg et al., 1997) and larger uncertainties in remote sensing retrievals (e.g,. . Direct measurements of aerosol absorption can be obtained by using instruments that measure scattering and extinction coefficients. Such measurements are limited to discrete 6 Aqua-MODIS radiances (at 466 and 646 nm) provided in the 10-km aerosol product (MYD04_L2) from the Deep-Blue (DB) aerosol algorithm. This aerosol product provides cloud-free radiances and ancillary information on the quality of pixel and estimated cloud fraction.

Methodology
A schematic flowchart of the method adopted in this work to derive wavelength-dependent aerosol absorption is 150 shown in Figure 2.

Computation of site-specific Look-up table of TOA reflectances
To start, we compile a seasonal climatology of aerosol particle size distributions and real part of the refractive index (440 nm) from the AERONET Level-2 Version-2 inversion product for each site considered in the study.
Here, we assume that the spectral variability of the real part of the aerosol refractive index through UV-Visible is 155 minimal and, therefore, values derived at 440 nm were assumed to be wavelength-independent across the UV-Visible spectrum range considered in this study. The resulting climatology of aerosol size distribution are fed to a radiative transfer model (RTM) to generate look up table (LUT's) of outgoing top of the atmosphere (TOA) reflectances at 340,354,388,466, and 646 nm with varying nodal points of satellite-sun geometry, surface pressure, τ, ALH, and imaginary component of the refractive index. The Gauss-Seidel radiative transfer code 160 (Mie theory) used for this purpose accounts for gaseous absorption, molecular and aerosol multiple scattering (Herman and Browning, 1965). Thus, a database of AERONET site-specific seasonal LUT of reflectances for the aerosols observed over each site in the study is created. Figure 3 shows the calculated net aerosol reflectance at the TOA over the GSFC site (38.92 o N, 76.84 o W) using particle sizes derived from the AERONET product and varying values of τ and ω o . These results illustrate that for a given satellite-sun geometry and observed radiance, 165 multiple combinations of τ and ω o can explain the satellite measurements. This simulation demonstrates that in order to derive ω o from satellite measurements, an accurate characterization of τ, cloud-free radiances, and surface reflectances are required.
The site-specific LUTs developed here assume spherical particle shapes for carbonaceous and urban aerosols. 170 However, mineral dust particles are assumed non-spherical and modeled as randomly oriented spheroid (Dubovik et al., 2006;Torres et al., 2018). To account for the non-spherical behavior of dust particles, a unified dust model is created using particle sizes from selected AERONET sites over Sahara and Arabian region that include: Saada, https://doi.org/10.5194/amt-2021-8 Preprint. Discussion started: 19 January 2021 c Author(s) 2021. CC BY 4.0 License. 7 SEDE_BOKER, Solar_Village, and Tamanrasset_INM. These sites were selected based on the observed prevailing dust aerosol type. The particle sizes and real refractive index obtained at these sites are used with a 175 pre-computed set of kernels that assume a spheroidal shape with a fixed distribution of axis ratio to produce phase function (Dubovik et al., 2006). The obtained phase matrix elements are input to the RTM to create reflectance LUT's. The process of acquiring a unified dust model LUT is necessary to account for the nonspherical shape of particles and save a considerable computational time, which otherwise would require to create another set of site-specific LUTs.

Collocation of satellite and ground measurements
We use satellite measurements located within the 50 km radius of each AERONET site. In essence, we treat the overlying atmospheric aerosols within a 50 km radius of the site as a representative of the AERONET measured τ. To allow for more sampling, we associate the AERONET observations within ±2 hours of satellite overpass to the measured TOA radiances. Here, we do not employ any averaging scheme for the AERONET data and keep it 185 intact. While with satellite measurements, we use native pixel resolution of 13 km x 24 km for the OMI wavelengths (340, 354 and 388 nm) and 10 km x 10 km resolution radiances for the MODIS wavelengths (466 and 646 nm).

Retrieval of aerosol ω o (λ)
The proposed technique to derive aerosol absorption follows the procedure of obtaining the best quality assured 190 cloud free-TOA reflectances, identifying the aerosol type, optimal layer height, and characterize surface reflectance. We select over-land pixels from both sensors with the best quality flags ('0'-OMI, '3'-MODIS-DB) and cloud fraction < 0.2 in the retrieval procedure. Aerosol type information for the OMI wavelengths is directly adopted from the OMAERUV product. While for the MODIS wavelengths, our algorithm looks for the nearest OMI footprint to obtain and assign the corresponding aerosol type. Once an absorbing aerosol type i.e., 195 carbonaceous smoke or mineral dust is identified, we choose the best estimate of ALH from the joint OMI-CALIOP climatology derived from a 30-month long record of collocated observations (Torres et al., 2013).
While for a weakly absorbing aerosol (Urban), ALH is characterized with a Gaussian distribution of aerosols with a peak at the surface. This is similar to the procedure adopted in the OMAERUV aerosol retrieval (Torres et al., 2013). For the surface characterization at OMI wavelengths, we use a near-UV surface albedo database used 200 in the OMAERUV algorithm. At MODIS wavelengths, surface reflectance provided by MAIAC products (Lyapustin and Wang, 2018) is used. Our retrieval technique gathers all above-mentioned information for each https://doi.org/10.5194/amt-2021-8 Preprint. Discussion started: 19 January 2021 c Author(s) 2021. CC BY 4.0 License. 8 pixel along with the associated AERONET τ to perform an inversion for each wavelength independently. The inversion procedure solves for the best fit of radiances and τ with the prior computed site-specific seasonal LUT radiances to derive aerosol ω o (λ). 205 Figure 4 shows the retrieved aerosol SSA over the GSFC site as a function of measured τ. Located in the vicinity of a metro city, the prevailing aerosols over the GSFC site are the urban or industrial types that are relatively more scattering in nature. The mean aerosol SSA retrieved at the GSFC site for all τ observations at 340, 354, 388, 466 and 646 nm are 0.91, 0.93, 0.93, 0.90 and 0.85, respectively. The variability of the retrieved SSA is high 210 at lower aerosol loading for all wavelengths. Particularly notable is the high variability of retrieved SSA in most τ bins for the visible wavelengths (i.e., MODIS bands). This is due to the weaker aerosol signal strength for urban type aerosols at lower aerosol loading in the visible spectrum, where the measured TOA radiances are dominantly contributed by the underlying surface. Also shown in figure 4 is the number of collocated observations that were used in the inversion and the percent of observations for which SSA is retrieved. For about 12 years of the 215 satellite and ground collocated observations used here, it is clearly evident that the number of observations from OMI is less than MODIS observations. The difference in the number of collocated observations stems partly from the OMI row-anomaly, cloud contamination, and the coarser pixel resolution. The percent of SSA retrieved observations varies widely even within the corresponding sensor wavelengths (OMI: 340,354,388 nm and MODIS: 466,646 nm). At times depending on the surface albedo used, the computed net aerosol reflectance 220 might exceed the LUT limits and produce SSA values above one or less than the maximum absorption in the LUT. We avoid this by constraining our inversion procedure within the LUT limits and do not allow for any extrapolation of the inputs. However, this leads to the unequal number of retrieved observations within the sensor wavelengths. In other words, for a given observation within the OMI or MODIS sensor, it is possible to have aerosol SSA retrieved at one wavelength and no retrieval (out-of-bounds) at other wavelengths. Also, it is worth 225 mentioning that for few sites located along the coasts or in the islands (e.g., Mauna_Lao, Ascension_Island, Nauru), we were either unable to retrieve aerosol SSA or the number of days with retrieval is quite low. This is a consequence of OMI's large pixel size, where the satellite measured radiances are often contaminated by clouds and mixed-signal from the surface that are challenging to resolve and lead to out-of-bounds in the inversion. 230 To examine the seasonal variation of aerosol SSA and its spectral dependence, we create a subset of the data that includes observations for which aerosol SSA is retrieved for all the corresponding sensor wavelengths simultaneously (OMI: 340, 354, 388 nm and MODIS: 466, 646 nm) on any given day. This step reduces the https://doi.org/10.5194/amt-2021-8 Preprint. Discussion started: 19 January 2021 c Author(s) 2021. CC BY 4.0 License. 9 sample size drastically but eliminates the need for making prior assumptions on the wavelength dependence of aerosol absorption angstrom exponent to fill those gaps. Instead, the obtained aerosol SSA in the UV-Visible 235 range is used to compute the resulting spectral dependence of aerosol absorption of the prevailing aerosols over the corresponding AERONET sites in terms of the Aerosol Absorption Exponent (AAE), a measure of the spectral dependence of aerosol absorption (Bond, 2001) using a power-law approximation, analogous to the Angstrom Extinction Exponent (van de Hulst, 1957). The spectral dependence of aerosol absorption is reported as Absorption Angstrom Exponent (AAE) defined as the slope of aerosol absorption optical depth with 240 wavelengths on a log-log scale. The aerosol absorption optical depth τ abs (λ) is derived as shown in equation (1): from which the AAE for wavelength range λ 1 , λ 2 is calculated as shown in equation (2).
In addition, we make use of the AERONET extinction angstrom exponent (α) at 440-870 nm to distinguish the 245 particles as coarse (α < 0.2), fine (α > 1.2), and in between as intermediate or mixed-mode. Since our aerosol identification strictly uses three primary types, the use of qualitative indicator α helps delineate the mixture of aerosols when applicable.

SSA retrieval sensitivity analysis
The inversion procedure employed here to derive aerosol absorption from the combined ground and satellite 250 measurements is susceptible to several systematic and random errors. These error sources include: (a) aerosol extinction measurements, (b) estimation of particle size distribution, (c) real part of refractive index, (d) calibration of satellite measured TOA radiances, (e) surface reflectance, (f) aerosol layer height, and (g) sub-pixel cloud contamination. The retrieved aerosol absorption from our inversion procedure could be affected by all these sources of uncertainties. Among these, error sources from (a-d) are inevitable for which we do not have any 255 direct control over them. However, we do have control only for the sources from (e-g) in our retrieval. Errors associated with surface reflectance, aerosol layer height, and cloud contaminations on the satellite retrieved optical depths are well documented in the literature (e.g., Fraser and Kaufman, 1985;Torres et al., 1998;Jethva et al., 2014). In summary it is known that an: (i) overestimation (underestimation) of surface reflectance leads to lower (higher) aerosol SSA, (ii) overestimation (underestimation) of τ leads to lower (higher) aerosol SSA, (iii) 260 overestimation (underestimation) of ALH produces higher (lower) aerosol SSA -more pronounced in the UV than in visible wavelengths, and (iv) an increase in TOA reflectance due to sub-pixel cloud contamination produces higher aerosol SSA. We use sensitivity tests for these key input variables to derive a quantitative estimate of the error percolated in the aerosol SSA retrieval due to changes in these variables. 265 Sensitivity tests are performed on observations that were reported with best accuracy (minimal cloud contamination in both OMI and MODIS data sets) for a few selected sites that include GSFC, Avignon, Tamanrasset_INM, Saada, Alta_Floresta, and Mongu. These sites were selected to include different types of aerosols observed over these sites. Figure 5 shows the error analysis of the retrieved SSA as a function of wavelength and optical depth given a change in the surface reflectance and ALH by an absolute change of ±0.01 270 and ±1 km ALH. The absolute error is computed as the SSA obtained with altered input minus the actual SSA.
As expected, the error in retrieved SSA increases with increasing wavelength and decreasing τ due to changes in surface reflectance for all aerosol types. For less-absorbing (Urban) aerosols, the surface reflectance becomes increasingly important at the visible wavelengths than compared to absorbing aerosols. Our analysis shows that for small τ 440 (~0.2), the error in retrieved SSA is much higher (> ±0.05) for visible wavelengths, while that in the 275 near-UV region reaches up to ±0.03. However, for observations with τ 440 ~0.4, a reliable accuracy of within ±0.05 is achieved even for the non-absorbing aerosols at 646 nm. In contrast to the surface reflectivity, the effect of ALH becomes prominent at near-UV wavelengths. The error in retrieved SSA due to changes in ALH decreases with wavelength because of the gradually diminishing the intensity of Rayleigh scattering and its radiative interactions with aerosols. The errors in the retrieved SSA are estimated to be better than ±0.03 in the 280 near-UV and negligible at visible wavelengths for both absorbing and non-absorbing aerosols.

Results
The results presented hereafter include only a data subset that meets the following three conditions: (a) SSA  (Table 3) reveals high absorption during JJA at CUIABA and Sao Paulo. It is likely that the urban aerosol samples shown here attribute to a mixture of aerosols.

Southern Africa
The AERONET sites located in the Southern Africa are Mongu, Pretoria, and Skukuza. In addition to the natural forest fires and emissions from crop residue burning, heavy industrial facilities and episodic dust commonly 315 dictate the aerosol amounts over Southern Africa. Figure 6b  is observed for DJF where the average SSA is almost flat from 340 nm to 466 nm with a slight increase thereafter. Aerosols organic components are likely the cause of such absorption spectral feature.

Australia
In general, inland Australia is categorized as arid-region vastly covered with deserts. However, Northern and Western parts of the continent are covered with savanna grasslands, where biomass burning due to natural forest 330 fires and land management practices are known to produce high aerosol emissions during the dry season (May-October). The regional average SSA for the aerosols observed over Northern Australia at the sites Jabiru and

Sahara
The seasonal average SSA of aerosols over the sites Saada and Tamanrasset is shown in figure 7a. has relatively low SSA (0.95) at 646 nm and therefore low UV-Vis AAE (2.1 to 2.5). This is consistent with the known dependence of scattering effects at longer wavelength for the dust particles.

Sahel
The AERONET sites located in the Sahel region are Agoufou, Banizoumbou, Dakar, IER_Cinzana, Ilorin, Ouagauodu, and Zinder_Airport. Regional average SSA for the aerosols over Sahel region is shown in figure 7b. 355 For dust aerosols, the average spectral SSA resembles typical dust absorption curve (increase in SSA with wavelength number of gas flaring stations (> 300) around the Niger Delta produces high emissions (Onyeuwaoma et al., 2015). Highly absorbing black carbon amounts observed at Ilorin is possibly a result of such emissions.

Arabian Peninsula
The AERONET sites located in the Middle East/Arabian Peninsula region include Solar_Village, SEDE_BOKER, Nes_Ziona, and Cairo_EMA_2. The regional average SSA of aerosols observed for these sites is 375 shown in figure 7c. For dust aerosols, as expected, the average SSA increases with increasing wavelength. The regional average SSA ranges from 0.89 to 0.98 at 340 nm to 646 nm, with UV-Vis AAE in the range of 2.7 to 3.8. No distinct seasonality in SSA is found from our sample of observations. However, a slight increase in SSA

400
Similar to the western part, atmospheric aerosols found over the eastern North America primarily originated from the industrial activities and secondary aerosol processes (Malm, 1992). Biomass burning generated carbonaceous particles and dust or mixture of aerosols over the eastern parts of the continent is a rare occurrence, except in the events of long-range transport of smoke from the west. Thus, the average aerosol SSA retrieved over the AERONET sites in this region follows a typical 'Urban' spectral absorption curve, as shown in figure 8b. It is 405 observed that aerosol SSA increases from the 340 nm to 388 nm or 466 nm and then decreases attaining a maximum absorption (~0.90) at 646 nm. The regional average SSA for the MAM and JJA months at 466 nm is 0.89 and 0.87 respectively. It should be noted that for urban aerosols the retrieval error at visible wavelength is high (up to 0.05). Since there are no notable changes in the aerosol sources, the observed decrease absorption at 646 nm for JJA period is likely attributed from the retrieval uncertainty. 410

Europe
For the AERONET sites located in the Europe, dominantly urban aerosols are observed (figure 8c). The regional average SSA increases from 340 nm to 388 nm and then decreases reaching a minimum value ( (Table 3) and winter (Table S1). While the aerosol loading is similar throughout the seasons, the average UV-Vis AAE for urban aerosols range from 1.0 to 420 1.26. In addition to the urban/industrial aerosols, long-range transport of dust from Sahara is not uncommon over of aerosols from fossil fuel combustion throughout the year, Mexico City also experiences biomass-burning aerosols during the relatively dry months of March-May from local sources (Eck et al., 1998).

North-Eastern Asia
The AERONET sites located in the Northeastern Asia include: Beijing, Osaka, Shirahama and XiangHe. Figure   440 9b shows the regional average SSA derived over Northeastern China. For the samples obtained, dust aerosols possibly mixed with regional pollution are observed over Beijing and XiangHe during spring (MAM waste burning. The spectral behavior of urban aerosols is similar to carbonaceous aerosols with decrease in magnitude of average SSA, AOD, and UV-Vis AAE. As expected from the local sources, mixture of aerosols categorized by particle sizes (0.2 < α 440-870 < 1.2) is observed at all sites in the region with widely varying spectral dependence.

455
The AERONET sites located in Northern India include Jaipur, Gandhi College and Kanpur. Major source of aerosols over the region includes industrial and vehicular emissions, combustion of biomass and fossil fuels. In addition, desert dust passage from arid and semiarid regions of northwestern India, Pakistan, and Arabian Peninsula is commonly observed during spring and summer months. Figure 9c shows the regional average aerosol absorption observed at Kanpur.

Discussion
Through extensive studies in the literature it is known that optical properties of biomass burning aerosols depend 480 on fuel/vegetation type, combustion processes, available moisture content (e.g., Ward, 1992;Reid and Hobbs, 1998;Reid et al., 1998;Eck et al., 2001). flaming combustion producing high amounts of soot, as also noted in our retrievals. Figure 10 shows the regional average AAE obtained for carbonaceous aerosols at three wavelength pairs. Overall, the average slope of absorption in visible (AAE 466_646 ) and UV-Vis (AAE 340_646 ) for carbonaceous aerosols is found to be within 2.
This is consistent with the studies that report AAE of biomass burning aerosols from several field campaigns in 500 the range 1 to 3 (Kirchstetter et al., 2004;Schnaiter et al., 2005;Bergstrom et al., 2007;Clarke et al., 2007).
However, the average AAE 354_388 obtained is high up to 4 for most regions. This is likely a result of higher organic matter in the regional biomass types and highlights the importance of UV spectral region in delineating Arabian Peninsula varies significantly. Figure 11 shows the regional average AAE obtained for dust aerosols at three-wavelength pairs. Among the dust-prone regions considered here, the regional average of the UV-Vis spectral dependence (AAE 340_646 ) is found to be close to or greater than 3 for all, except the Sahel and Eastern China, where average value ranges 1.5 to 2.5. Although no distinct seasonal variation in spectral absorption of dust is noted, the variability in spectral dependence over the regions is quite evident. Overall the regional average of AAE for dust aerosols observed here is consistent with insitu measurements that report values ranging 1.5 to 3.5 (Bergstrom et al., 2004(Bergstrom et al., , 2007Müller et al., 2009;Petzold et al., 2009). However, observations at individual sites (Table 3) show that the spectral dependence of the observed dust for few sites is relatively high than those 530 reported by insitu measurements. Considering our retrieval method where aerosol absorption is derived independently for each wavelength and have computed the dependence, our results agree reasonably well with the insitu measurements reported in the literature.
While urban aerosols constitute dominantly sulfates and other forms of nitrate particles, industrial emissions and 535 fossil fuel combustion produces various forms of carbon that contribute to the overall optical properties. Further the aerosol size growth due to increase in relative humidity in the atmosphere and coagulation processes are known to alter the absorbing nature of aerosols. Figure 12 shows the regional average AAE obtained for urban aerosols at three-wavelength pairs. Urban aerosols in highly polluted environments such as over the Mexico City have near unity spectral dependence. While the passage of biomass burning emissions over such environment 540 show unusual decrease in absorption at the UV region attributing to high AAE 354_388 . This is consistent with studies that report relatively high AAE in UV region and near unity in visible region for the aerosol mixture consisting of organic matter and black carbon amounts (Barnard et al., 2008;Vanderlei Martins et al., 2009;Bergstrom et al., 2010;Jethva and Torres, 2011). The urban aerosols found in our sample over Northern Indian and Eastern China are highly absorbing exhibiting AAE 340_646 ~1.5 than the carbonaceous aerosols with 545 AAE 340_646 ~2. These results suggest the combination of magnitude of aerosol absorption and its spectral dependence in UV, visible and UV-Visible spectrum could be used to partition mixture of aerosol types found in such environments. Overall the regional average UV-Visible AAE for the urban aerosols is found to be near 2.

Comparison with AERONET SSA product
We compare of our aerosol SSA retrievals at the visible wavelengths with that available from AERONET data   Figure 15a shows the comparison of retrieved SSA with AERONET by combining all aerosol types together. Our retrieved SSA at 466 and 646 nm agree with AERONET SSA for 34% (0.06) and 40% (0.07) of observations (RMSE) respectively. The absolute difference in retrieved and AERONET SSA as a function of optical depth is shown in figure 15b. The differences in SSA for both wavelengths at 466 nm and 646 nm are higher for lower τ and become negligible for higher τ. It is observed that at 466 nm, the observations with positive differences are 595 relatively more than that at 646 nm. It is important to note here that our retrieval method and that used in the AERONET inversion differ fundamentally in several aspects. Other than the different source of surface reflectance data used, one major distinction between the two methods is that the present inversion algorithm retrieves SSA at different UV and Vis wavelengths independently, whereas the AERONET algorithm internally applies a condition ensuring the spectral shape of SSA follows an expected pattern for the observed aerosol type 600 (private communication with Thomas Eck, NASA GSFC). The differences noted between our SSA retrievals and that from AERONET at different wavelengths could be attributed, at least partially, to this treatment of reported values of SSA.

Summary
Ground-based measurements of direct solar radiation under cloud-free conditions over worldwide sites are 605 providing valuable insights into regional aerosol characteristics. Long-term measurements obtained from such network, such as from AERONET, are widely used to develop regional aerosol climatology and investigate seasonal/annual variability. Satellite measurements of TOA radiances are able to provide global distribution of columnar aerosol amounts. However, deriving aerosol optical properties from satellite measurements require constraints on particle sizes and optical properties. Reliable aerosol measurements from ground-networks and 610 airborne/field campaigns are traditionally used to validate and improve the constraints in satellite aerosol retrievals. In this work, we use AERONET measured extinction τ as constraint in a robust inversion technique that uses satellite measured TOA radiances from OMI and MODIS to derive spectral aerosol absorption in the UV-Vis part of the spectrum. Other than cloud contamination of the TOA radiances, major sources of error in our biased toward dense pollution/industrial, smoke, and dust events, less-absorbing low aerosol amounts seldom have dependency on the ω o . Therefore, the regional aerosol absorption models derived here offer essential guidance for selecting spectral absorption in satellite aerosol retrievals spanning the UV-Vis spectrum. From our analysis of worldwide inland sites: (a) it is suggested that satellite aerosol retrieval techniques could employ 680 regional dynamic absorption models to avoid potential bias in τ retrievals noted in earlier studies, and (b) the spectral dependence of aerosol absorption noted here for the UV (354-388 nm), visible (466-646 nm) and UV-Visible (340-646 nm) range for all aerosol types other than black carbon varies considerably. Overall, the UV absorption data set well compliments and provides more information on the regional aerosol absorption than with the visible data set alone. 685 Given the lack of aerosol absorption information at near-UV wavelengths in the existing AERONET record and limited availability of insitu measurements, the UV-Vis aerosol absorption data set developed here, perhaps for the first time, offers a valuable source of information useful for a variety of aerosol and trace gas studies. The analysis presented here focuses on regional aerosol absorption using a subset of results. The derived spectral 690 dependency can be used with all SSA retrievals to construct and investigate long-term trends in UV-Visible aerosol absorption. Further, the spectral aerosol SSA derived here could be used to parameterize absorption in models and better understand the radiative effect of aerosols. Our ongoing investigation utilizing the complete data set developed here will explore some of these applications in the future.

Author Contributions
Omar Torres (OT) and Hiren Jethva (HJ) had conceptualized the research. Vinay Kayetha (VK) developed the data set, performed formal analysis, and wrote the manuscript with inputs from OT and HJ. All authors reviewed results, helped with the data interpretation and edited the manuscript to make a final version.

Data Availability
The spectral aerosol absorption data set developed here will be made available upon request to the authors.