Comparison of UV irradiances from Aura/Ozone Monitoring Instrument (OMI) with Brewer measurements at El Arenosillo (Spain) – Part 2: Analysis of site aerosol inﬂuence

Several validation studies have shown a notable overestimation of the clear sky ultraviolet (UV) irradiance at the Earth’s surface derived from satellite sensors such as the Total Ozone Mapping Spectrometer (TOMS) and the Ozone Monitoring Instrument (OMI) with respect to ground-based UV data at many locations. Most of this positive 5 bias is attributed to boundary layer aerosol absorption that is not accounted for in the TOMS/OMI operational UV algorithm. Therefore, the main objective of this study is to analyse the aerosol e ﬀ ect on the bias between OMI erythemal UV irradiance (UVER) and spectral UV (305 nm, 310 nm and 324 nm) surface irradiances and ground-based Brewer spectroradiometer measurements from October 2004 to December 2008 at 10 El Arenosillo station, with meteorological conditions representative of the south-west of Spain. The e ﬀ ect of other factors as clouds, ozone and the solar elevation over this intercomparison were analysed in detail in a companion paper (Ant´on 2010). In that paper the aerosol e ﬀ ects were analysed making only a rough evaluation based on 15 aerosol optical depth (AOD) information at 440 nm wavelength (visible range) without applying any correction. Here, the correction of the OMI UV data is proposed based on a detailed study about the determination of absorbing aerosols provided by AERONET data. Because of the di ﬃ culty to have reliable data about absorbing aerosol properties at many sites, ﬁrst we consider AOD, Angstrom exponent and also OMI-Aerosol


Introduction
It is well established by several validation works (Arola et al., 2005(Arola et al., , 2009Antón et al., 2007;Kazantzidis et al., 2006;Kazadzis et al., 2009) that UV irradiance at the ground given by satellite instruments such as TOMS (Total Ozone Mapping Spectrometer) and OMI (Ozone Monitoring Instrument) (Krotkov et al., 2002) is overestimated compared with ground based UV measurements at many polluted locations. This positive (satel- 15 lite overestimation) clear-sky bias varies widely depending on the climatological conditions at validation sites. Most of this positive bias is attributed to boundary layer aerosol absorption that is not accounted for in the current TOMS/OMI algorithm (Krotkov et al., 2005;Arola et al., 2009).
The OMI instrument is a Dutch-Finnish push-broom UV-Visible spectrometer, a con-correction is based on a global climatology of aerosols, using combined information of aerosol models and AERONET data. This work has been carried out over seven European sites (including El Arenosillo station) and showed a reduction of the bias from 22 to 17% for the UV spectral irradiance at 324 nm under different atmospheric conditions. In this framework, our work is focused on the comparison between OMI UV irradi- 15 ance products (erythemal irradiance (UVER), and spectral irradiances at 305, 310 and 324 nm) and the ground-based UV measurements using Brewer spectroradiometer located at El Arenosillo (Huelva), representative of the SW of Spain. In a companion paper (Antón et al., 2010) the effect of several factors as clouds, aerosols, ozone and the solar elevation on OMI-Brewer UV bias were analysed and compared with pre-20 vious TOMS-Brewer results (Antón et al., 2007). In the previous works, the aerosol effect was analysed in terms of aerosol extinction optical depth (AOD) measured at visible 440 nm wavelength. The comparison results showed that under moderate-high aerosol load (AOD>0.25) the OMI bias is about 19-15% for UVER and spectral UV irradiances. Under cloud-free and low aerosol load conditions (AOD<0.1) the OMI bias 25 was smaller ∼11% for UVER with similar spectral results. These studies had not attempted correcting OMI UV data for aerosol absorption, because no information about aerosol absorption properties at UV wavelengths was available at our site.

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | In this paper we first analyse absorption properties of local aerosols at El Arenosillo site, representative area of South Europe, summarizing our previous measurements (Toledano et al., 2007a(Toledano et al., ,b, 2009Cachorro et al., 2006Cachorro et al., , 2008Prats, 2009). Even discrimination of absorbing and non-absorbing aerosols is challenging task, related to the discrimination of aerosol types. Measurements of absorption aerosol optical thick-5 ness (AAOT) or single scattering albedo (SSA) is routinely available only in the visible wavelengths (440 nm and longer wavelengths, Dubovik et al., 2002). The uncertainty of these parameters from AERONET inversions is not yet known. Furthermore, the spectral dependence of SSA depends on aerosol type, thus, AERONET SSA values at visible wavelengths can not be simply extrapolated into UV wavelengths (Krotkov et al., 10 2005(Krotkov et al., 10 , 2009Arola et al., 2009).
El Arenosillo is an appropriate location for aerosol studies because of a high frequency of cloud-free days, and a great variety of aerosol types with a high occurrence of desert dust outbreaks from Africa during all seasons (Toledano et al., 2007a(Toledano et al., ,b, 2009). In addition, this location enjoys availability of long term measurements of differ-15 ent aerosol types. The period of study is determined by availability of OMI data from October 2004 to December 2008.
The paper is organized as follows. Site characteristics, ground and satellite-based measurements are briefly described in Sect. 2. Section 3 introduces the methodology used for comparison between OMI and Brewer measurements. The comparison results 20 are presented in Sect. 4 with focus on different approaches for OMI aerosol correction. Finally, Sect. 5 summarizes the main conclusions.

Site, instruments and data
Short comments about site, instruments and the type of data are given in this paper. For a more detailed description read the earlier mentioned companion paper (Antón belongs to the Earth Observation, Remote Sensing and Atmosphere Department, National Institute of Aerospace Technology of Spain (INTA). This centre participates in the Global Ozone Observing System (GO3OS) of the Global Atmosphere Watch (GAW) program of World Meteorological Organization (WMO) as station #213. Data gathering, retrieval and reporting procedures at these stations are standardized by the WMO 10 quality assurance procedures.
Aerosol data are provided by the CIMEL-AERONET sun-photometer available since year (Toledano et al., 2007a,b, 2009. Aerosol optical depth AOD or thickness AOT (we use both indistinctly) and the derivedÅngström coefficient (also denoted by alpha) are obtained from direct sun measurements using the current version 2 of 15 AERONET algorithm or by the GOA algorithm (Cachorro et al., 1987(Cachorro et al., , 2001Ortiz de Galisteo et al., 2009). The absorption aerosol optical depth or thickness, AAOD (AAOT) and single scattering albedo (SSA) parameters are derived from the more sophisticated inversion algorithm used by AERONET.
The UV irradiance weighted with the erythemal action spectrum adopted by the Com-20  Krotkov et al., 2002Krotkov et al., , 2005Tanskanen et al., 2007;Torres et al., 2007). In this study OMI UV products correspond to the new version of the OMI level 1 (radiance and irradiance) and level 2 (atmospheric data products) data set named collection 3. This new version takes advantage of a coherent calibration and revised dark current correction. For more information, please, 5 visit the NASA DISC at http://disc.gsfc.nasa.gov/Aura/OMI/ for EOS Aura OMI level 2 orbit data, and the Aura Validation data Center at http://avdc.gsfc.nasa.gov for EOS-Aura OMI station overpass data. Please consult the OMI README files for the latest OMI data product information.
For the latitude of El Arenosillo station, OMI instrument provides more than one 10 overpass per day. In this work, we used daily OMI data recorded closest to the groundbased location. Thus, the distance between the centre of the satellite pixel and the station vary from 0.1 km to 48 km, with a average value of 11.5 km. In addition, Brewer data recorded closest to the daily OMI overpass time (between 12:30 and 14:30 LST for El Arenosillo station) were used. The average of the daily differences between the 15 time of Brewer measurements and the OMI overpass time is 6 min. The OMI-Brewer data with time differences higher than 15 min (∼5% of all data) were removed in the comparison. In a previous paper by Antón et al. (2010), OMI-Brewer irradiance under all-sky conditions with 1206 available data, representing 80% of total days during the period of 20 study, were compared. In that work, the OMI Lambertian Equivalent Reflectivity (LER) at 360 nm was used to describe cloud-free conditions. A certain day was classified as cloud-free if LER was lower than 10% (Kalliskota et al., 2000). Thus, 703 (49% of the total days) days were selected as cloud-free. In this study, CIMEL AERONET cloud screening data have been also used to define cloud-free conditions, due to the need 25 of simultaneous AOD and OMI-Brewer irradiance data. Thus, the number of selected data is slightly lower, 633, and finally reduced to 583 (45% of the total days) in order to use the available SSA values, as we explain below. This high number of values (as was shown in Arola et al., 2009)  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | lar radiation and aerosol studies, currently providing of one of the best ten AERONET long-continuous data-set, suitable to analyse the OMI bias due to aerosols.

Methodology
The comparison analysis for irradiance is carried out by regression analysis and also by the mean bias error, MBE, and the mean absolute bias error MABE. These parameters 5 are defined by: The uncertainty of MBE and MABE is characterized by the standard error (SE). These two statistical parameters were used according to the methodology used in previous 10 works ( Antón et al., 2007Antón et al., , 2010. For the OMI UV data correction the formula provided by Krotkov et al. (2005) based on the aerosol absorbing correction factor (CAA) was used: where λ is the corresponding OMI working wavelength in nm. CAA can be estimated 15 by the aerosol absorption optical thickness (AAOT) as follows: The denominator describes the linear dependence of the overestimation ratio of satellite-based UV on AAOT being b the constant slope. This parameter weakly depends on solar zenith angle (SZA) and aerosol type (Arola et al., 2005; Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2005). A value of b=3 has been used as the most current value. However, according to its definition and following the approach of Kazadzis et al. (2009), the parameter b must be determined from the available data. Although CAA will be applied to all OMI UV products, the wavelength λ is restricted to 340 nm, the only available for AAOT. Therefore, AOT and AAOT (or the equivalent SSA) at UV wavelengths are necessary 5 for OMI correction data. AAOT (or SSA) needs more sophisticated inversion algorithms for its determination, presenting a high uncertainty when available. These parameters are retrieved at visible-NIR spectral range by the AERONET inversion algorithm. This algorithm is based on aureole radiance data, restricting SZA to low values and with others limitations, such as AOD greater than 0.4. All these restrictions considerably reduce the number of available site-data. In the UV range, AOT values are more scarce than those at visible or near infrared in AERONET stations. In addition, there is no current retrieval methodology available with AERONET data for AAOT or SSA in the UV range. To overcome this issue, Krotkov et al. (2005) use UV-MFRSR radiometer aerosol data, and in the work of Kazadzis et al. (2009) all the aerosol information is 15 obtained from Brewer measurements. However, these methodologies present some limitations for aerosol retrieval radiative properties. SSA is a function of the wavelength and this dependence can be used to define the aerosol type. Most current discrimination of aerosol types is based on AOD-alpha plots information Toledano et al., 2007a) which may give additional 20 information to select partially absorbing aerosols, but not to retrieve reliable absorbing properties. Therefore, here different approaches will be followed, from AERONET aerosol information based only on AOD-alpha to a more detailed approaches based on SSA or AAOT.
Moreover, Aerosol index AI (Hsu et al., 1999;Torres et al., 1999) provided by the OMI sensor also gives information about absorbing aerosols. In this study, this approach has been explored in correspondence with AOD-alpha or SSA values in order to get more detailed aerosol information. Introduction For this analysis the results of our previous paper of Antón et al. (2010) must be borne in mind. The overestimation of OMI irradiance values about 13% for cloud-free conditions and the initial evaluation of aerosol influence were very informative. Nevertheless, that paper did not analyse the influence of absorbing and non-absorbing aerosols on 5 the observed OMI bias and hence no correction could be applied. It is needed to achieve a more detailed analysis of absorbing aerosols in order to correct its effect on OMI bias. This topic will be discussed in the next sections. However, the first issue is how to identify absorbing aerosols. As already mentioned different approaches have been considered depending on the available aerosol information.

Absorbing aerosol determination based on AOD-alpha and AI information
A first approach for determining absorbing aerosols based on AOD-alpha information was evaluated and its relationship with OMI bias was investigated. The most common and widely available information about aerosol properties are AOD and alpha parameters. Thus, climatological tables for both parameters are provided by AERONET. 15 However, this is not the case for SSA.
Values of AOD at UV wavelengths are needed to perform this analysis. The standard sun-photometer installed at El Arenosillo has no measurements at UV wavelengths since it is a polarized model. In order to overcome this drawback, AOD at 340 nm was measured by another Cimel photometer which was installed at our station during more 20 than a year (between June 2006 and August 2007) within the period of study. Thus, AOD values at 440 nm and the alpha parameters have been used to extrapolate the AOD to UV wavelengths. This approach has been validated using the measurements of AOD values at 340 nm. In this sense, Fig. 1a shows a good agreement between the measured and calculated values at 340 nm.
It would be desirable to have AOD values at shorter UV wavelength, lower than 340 nm, but only Cimel data at this wavelength are available at this moment. AOD Brewer data were not used in this work since there is currently a high disagreement between our retrieved Brewer data and Cimel data at UV wavelengths. Although many works give a good assessment about AOD Brewer retrieval (Gröbner et al., 2001;Marenco et al., 2002;Gröbner and Meleti, 2004;Kazadzis et al., 2005Kazadzis et al., , 2007Cachorro et al., 2009) some problems remain that must be solved to have reliable Brewer 5 AOD data in an operational way, without additional measurements as in the works by Kazadzis et al. (2005Kazadzis et al. ( , 2007. Further improvements are necessary in calibration constants to obtain operational AOD Brewer data. It is interesting to analyse the relationship between AOD at 340 nm and AOD at 440 nm ( Fig. 1b) since part of the information given at UV range is derived from vis-10 ible spectral range and in this case, by the alpha coefficient orÅngström exponent (hereafter we use alpha). The figure shows two branches: one corresponding to the black points and the other to the grey points. This behaviour could disagree with the expected more linear correlation between both wavelengths (or between others pair of wavelengths: i.e., 440-670, etc. . . ). Nevertheless, this behaviour appears when there and fine-moderate size particles. The peak of the coarse mode, about alpha=0.5, is mainly due to desert dust aerosol type and the other mode (about alpha=1.2) is mainly related to the prevailing coastal marine aerosols. Fine particles have a low weight in the histogram for our station but, in the latter mode, continental aerosol type and fine particles associated with polluted or other type of events have been included. 25 Hence, the cloud-free data set have been divided into two groups, one of alpha values lower than 0.82 and another one with values equal or greater than 0.82 (representing the 28.3% and 71.7% of all cloud-free cases, respectively) according to the information given by Figs Fig. 1b when the value of AOD(440 nm) reaches 0.25. For AOD values lower than 0.25 both branches are joined, being difficult to separate both groups. Therefore, AOD(440 nm)=0.25 is considered as the limiting point between both branches. The additional information given by the alpha parameter allows us to perform this separation for lower AOD values. Figure 2b shows the relationship between 5 the alpha coefficient and AOD at 440 nm. Two data sets of AOD≥0.25 and one data set AOD<0.1 can be very clearly seen. These two groups of high AOD differentiate between moderate-fine (alpha greater than 0.82) related to polluted aerosol episodes and large particles (alpha lower than 0.82) associated with desert dust particles. Hence, it is expected that these two groups may represent absorbing aerosols.

10
It must be remembered that desert dust (DD) aerosols attenuate the UV radiation more strongly than other aerosol type with the same AOD due to its low single scattering albedo at UV wavelengths (Krotkov et al., 1998;Meloni et al., 2003). In the case of our area of study and taking into account their characteristics (Toledano et al., 2007a,b) polluted aerosols are represented by those arriving from industrial-urban ar-15 eas mixed with marine or continental aerosols, but not pure coastal-marine aerosols (AOD lower than 0.25). However, values of high alpha and AOD may also correspond to biomass-burning (BB) aerosols. Aerosols of this type correspond to isolated episodes well identified in our station (Cachorro et al., 2008). They are mainly related to cases of local forest fires and those arriving from Portugal during the period of study. BB 20 aerosols correspond to 5-6 points and, hence, represent a low percentage of the total number of cases (about 0.5-1%). Figure 3a shows the relationship between the relative differences OMI-Brewer and the valid associated values of AOD at 340 nm for cloud-free conditions (black rhomboid points). A very low correlation is found with R 2 =0.17 (R=0.41). This correlation is 25 notably improved when data with alpha below 0.82 are selected ( Ialongo et al., 2008;Buchard et al., 2008). A better correlation is found for binned data, as shown by Antón et al. (2007). Note that this correlation is relevant because is the correlation that works if we consider the case of SSA constant as we will discuss later in this paper. Although the correlation is very poor, the OMI bias increases as the AOD increase.

5
In order to gain more information about our selection of absorbing aerosols Fig. 3b shows the differences OMI-Brewer as function of AI. Similar to above, regressions were obtained showing very poor correlation values. However, according to AI values most points of the selected set of high AOD appear as non-absorbing (red points) with the exception of few of them with AI values greater than 0.5, corresponding to BB aerosols 10 (located close but at the right of the vertical line AI=0.5). We have taken AI=0.5 as the limiting value between absorbing and non-absorbing aerosol according to Krokov et al. (2005). The other set with AOD>0.25 and alpha<0.82 correspond to desert dust (DD) and are located at the top-right and appear as absorbing data.
To better classify absorbing aerosols we have plotted AOD(440 nm) versus AI in

20
AI is a nice way to detect absorbing aerosols, but since it is also sensitive to AOD and SSA quantitative information is more difficult to retrieve. Bear in mind that AI is very sensitive to the aerosol altitude and increasing with altitude, if other relevant factors, as AOD and SSA are the same, and not very sensitive to the aerosol boundary layer. Therefore, the information about red points may say more about the altitude than 25 about the strength of absorption. Red points may be associated with polluted boundary layer aerosols and also related to sulphates composition (Gonzalez-Castanedo et al., 2008) or more scattering aerosols. Because of this apparent inconsistency about absorbing and non-absorbing aerosols based on AOD-alpha and AI information, we get Introduction

Absorbing aerosol based on SSA values and its relation with AOD, alpha and AI
Two groups of high AOD with potential absorbing properties have been selected. It is worth noting that they only correspond to a small subset of the total free-cloud data 5 (15%). Figures 3 and 4 show the behaviour of these two subsets and also the whole data set in order to distinguish the different absorbing features of these data. Due to the non sufficient information with AI and AOD-alpha parameters, the available SSA values at our station given by AERONET must be considered, as the most relevant parameter to define the absorbing properties of aerosols.

10
Although AERONET provide SSA (or AAOT) values at the four aerosol wavelengths according to AERONET algorithm (Dubovik and King, 2000;Dubovik et al., 2002), a climatology is not usually feasible at most AERONET sites. Level 2 for SSA requires a set of restrictions which reduce considerably the data and hence it is difficult to have a representative climatology. Additionally, the restriction of AOD(440 nm)≥0.4 15 eliminates most of the aerosol data, as in the cases of our station.
All these conditions make it impossible to obtain a climatology based on AERONET level 2 data nor a reliable seasonal characterization at El Arenosillo in spite of using more than nine continuous years of data and taking into account its excellent weather conditions (Prats et al., 2008). Furthermore, the low number of data is not sufficient to 20 analyse OMI irradiance data and they are only representative of desert dust aerosols (Toledano et al., 2007b). Therefore it was necessary to take SSA level 1.5. Mean daily values (not overpass data) were used as the most reliable values for this analysis because of the restrictions of the inversion AERONET algorithm for SZA as mentioned above. The error of SSA reported by AERONET depends on the values of 25 AOD(440 nm) and also on aerosol types (DD, BB, etc.), with higher uncertainty for lower wavelengths (Dubovik et al., 2002), varying from 0.03 to 0.07. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | In spite of this assigned error, we must emphasize that the quality of the obtained SSA data-set at level 1.5 is not guaranteed (quality not assured according to AERONET). In addition, to check the SSA quality with other methodology is not generally feasible. This is the case for our analysis and unfortunately for other reported cases (Pére et al., 2009). To solve this problem, the work of Arola et al. (2009) pro-5 posed to use model data in combination to AERONET data in order to get a climatology expressing the SSA values as monthly-means. Also monthly SSA values were used in the work of Kazadzis et al. (2009) but using other methodology based on Brewer global-direct measurements and a radiative transfer model. For the moment at this station only AERONET data are available. Figure 5a shows SSA(440 nm) versus AOD, both taken at 440 nm wavelengths. A first look of this figure indicates that SSA values range from 0.7 to 1 with a 66% of points between 0.9 and 1. In addition, there are a significant number of points below 0.85 (14%), presenting most of then AOD(440 nm) smaller than 0.25. The cases with SSA(440 nm) smaller than 0.9 and AOD(440 nm) smaller than 0.25 represent 29% 15 of all. Our selected two groups of data with AOD≥0.25 appear as the top-right side, with most of them with SSA values between 0.9-1. Thus according SSA, they are not absorbent, as it was expected. The cloud-free data set has a mean value of SSA(440 nm)=0.9 and the most frequent value is 0.95 (histogram of frequencies not shown). Also it can be seen that the alpha values are more or less homogeneous 20 distributed around all the SSA range, not being associated with the absorbing characteristics.
In Fig. 5b SSA(440 nm) values are plotted versus AI for the whole data set and the set of alpha lower than 0.82 together with both sub-groups of AOD≥0.25 (red and green points). It can be seen that there is no correlation as it may be expected. Ap- 25 pearing at the left of AI=0.5 most of the points correspond to moderate-low absorption around SSA=0.92 and a non-neglected number of points show very low SSA values (high absorption) which correspond to AOD lower than 0.25. As before, red points (polluted) appear as non-absorbing according to AI but with a mean SSA value of 0.92, Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | thus as moderately absorbing. Green points or desert dust (DD) have a mean value of SSA=0.9 (lower than red points), giving also low-moderate or non pronounced absorbing characteristics. It must be empathized that these same cases appeared as absorbing aerosols according to AI (right side). In conclusion AI and SSA values do not show contradictory information to define absorbing aerosol groups. It is difficult to 5 choose what experimental values (AI or SSA) are most adequate to describe absorbing aerosol properties, although SSA parameters appear as the most reliable according to Mie theory.
For the OMI bias correction it is necessary to obtain the SSA at UV wavelengths where no retrieval methodology is currently available with AERONET data. In order to 10 solve this issue, we extrapolate from 440 nm to 340 nm assigning the same slope that SSA has between 440 and 670 nm as a good approach. Figure 6 shows the correlation between the SSA at 440 nm and at 340 nm. The agreement is high (R 2 =0.97), being the values at 340 nm slightly higher than at 440 nm. To illustrate the spectral behaviour of SSA Fig. 7 depicts the mean value of SSA at level 1.5 for the four aerosol wave- 15 lengths used in AERONET during the whole OMI analysed data set in El Arenosillo. Also it is depicted the SSA values of the whole data-set without desert dust (DD) group and the SSA of the DD group. As can be seen the SSA decreases for larger wavelength and the contrary happens for DD aerosols. Slightly lower values have been obtained for the period 2000-2008, as shown below in Fig. 8b. 20 The SSA values obtained in El Arenosillo are, in general, quite lower (more absorbing) than those found in other areas. Our values are equivalent or lower than those found in big urban cities as Paris or Mexico, or more polluted areas as the site of the GSFC (Goddard Space Flight Center) at Greenbelt near Washington-DC, as it is depicted in Fig. 8a. The three AERONET sites have a long-record of data as indicated in 25 Fig. 8a for level 1.5 and 2 where the observed difference are sufficient illustrative to put in discussion the true validity of data at level 1.5.
To confirm these non-expected high absorbing values of SSA of El Arenosillo we have also analysed the SSA values for our AERONET station of Palencia in the north Introduction of Spain. Although the site is located near a little city of 82 000 inhabitants without industry, very low values of SSA are not expected. In Fig. 8b the earlier values of El Arenosillo are depicted together with that of Palencia for level 1.5 and 2. SSA values in level 1.5 lower in Palencia than in El Arenosillo were obtained. It must be noted that level 2 is represented by the features of desert dust aerosols where SSA increases with 5 the wavelength, because in both sites only DD intrusions reach values of AOD as high as 0.4 to reach level 2. Therefore, this is a non expected result because El Arenosillo is located in a rural area and although it may have some influence of the industrial belt of the city of Huelva, this can not reach this lower SSA values. Observe that GSFC at level 2 has a value of SSA at 440 nm higher than 0.95 and at UV about 0.98, 10 thus appearing as a reliable value. On the other hand, the error of SSA reported by AERONET depends on the values of AOD(440 nm) and also on aerosol types (DD, BB, etc.), with higher uncertainty for lower wavelengths (Dubovik et al., 2002), varying from 0.03 to 0.07. The above results show the current state of the research about the values and un- 15 certainties of the SSA parameter. In spite of these shortcomings, the SSA values at level 1.5 will be used to analyse their relationship with the OMI irradiance correction.

Application of corrections for OMI bias. Approaches of SSA constant and SSA variable
The SSA of DD aerosols increase as the wavelengths increase (Dubovik et al., 2002) 20 in contrast with the behaviour of the SSA for the whole data set (with and without DD aerosols) which decreases. Because of this opposite spectral behaviour both sets have the same value of 0.915 for 340 nm, when the straight line that joins the SSA values at the wavelengths of 440 and 670 nm is extrapolated. We must also note that the spectral behaviour of the red points is very similar to the spectral behaviour of the whole data 25 set. Therefore, as first approach a fixed SSA value of 0.915 will be considered to generate the AAOD=(1+AOD×SSA) at 340 nm for the whole data set. In this case, the AAOD variation is exclusively due to AOD changes. As second approach, AAOD 16402 The two plots of Fig. 9 show the relationship between OMI/Brewer ratio for UVER and AAOD at 340 nm obtained by means of the first approach (top) and the second one (bottom). The fitting has been added to each plot in order to determine the parameter b of the correction formula for the two approaches. The value b=3.42 was obtained for the first approach which is near the proposed value of Krotkov et al. (2005) and b=1.155 for the second approach taking SSA(340 nm) variable. Both fits show very poor correlation with coefficient of determination R 2 equal to 0.17 and 0.04, respectively. The constants for the spectral irradiances at 305, 310 and 324 nm were also obtained and given in Table 1 (plot are not shown). Figure 10 shows the correlation between OMI and Brewer values for UVER before 15 and after the correction using the first approach (top) and the second one (bottom). agreement between satellite and ground-based irradiance data in our area, as has been reported previously (Antón et al., 2010). The correction of the OMI irradiance data (Eqs. 3 and 4) reduces significantly the bias with respect to Brewer measurements. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | creases to 11.65% (SSA variable approach) and 8.4% (SSA constant approach), which represents a relative decrease of 13% and 37% with respect to the values without correction. Similar improvement was obtained for the three spectral wavelengths. The smallest bias correspond to wavelengths at 310 nm (6.9%) and 324 nm (6.3%) for the SSA constant approach, representing a relative reduction of OMI bias of 32% and 38%, 5 respectively. As can be seen a better improvement has been obtained for the SSA constant approach which may be surprising in principle, but not taking into account the above discussion about SSA values. A fixed value of SSA equal to 0.915 at UV range is more reliable for the whole data set that the variable values of SSA at level 1.5. As it was 10 discussed above, AAOD values obtained for SSA variable appears as a whole as very absorbing, reaching a value of 0.17. The correlation between these two sets of AAOD values has been evaluated showing a low correlation, smaller than 0.3 (not shown). Thus, the constant slope b in the SSA variable approach decreases considerably, reducing the posterior effectiveness of the correction (see that b=1.155 is near 1). When 15 values greater than 0.08 are removed, a value of b=1.49 is obtained but the improvement is not significant.
Otherwise, if a value of b=3 is applied (Krotkov et al., 2005) the OMI bias of UVER decreases to 9.4% (324 nm falls to 4.5%) but the application of this value is not justified. Additionally, it is worth noting that for the second approach using SSA at 440 nm 20 instead of at 340 nm gives the same results. Also, the performance of the correction has been evaluated in the case of using a constant value of SSA equal to 0.95, which may be a reasonable value for the whole data set assuming less absorption, as in the area of GSFC (where the interpolate value go to 0.95 at UV range). In this case, the slope b increases up about 5 but the AAOD values decrease in such a way that the 25 correction factor, and hence the correction, are substantially modified. Thus, these other explored possibilities or approaches do not improve the previous one. Figure 11 depicts OMI bias as function of AOD(340 nm) before (same as earlier Fig. 4a) and after the application of the two corrections. Now no dependence is ob-served with horizontal slopes for the regression lines after the correction. According to the results of Taskanen et al. (2008), our results are inside the range to be considered as satisfactory. Therefore, these results indicated the level of accuracy that may be reached by the OMI algorithm at the present status of research. Otherwise, ongoing investigations must be accomplished to better assess SSA values and hence 5 to improve the estimated OMI values.

Conclusions
This work focuses on the influence of aerosols over the comparison between OMI and Brewer UV products at El Arenosillo station (south of Spain) for the period October 2004-December 2008. Under cloud-free conditions, our ground-based measurements 10 overestimate the OMI data, around 13% for UVER and ranging from 10% to 15% for the spectral wavelengths, in agreement with other locations. When the aerosol corrections are applied, we find an OMI bias ranging from 6-10% (SSA constant approach), which represents a relative reduction between 30% and 40% with respect to OMI bias without aerosol correction. For the SSA constant approach, the variation of AAOD depends 15 only on AOD features. Less effective reduction is obtained when using available mean daily SSA values, which in principle would appear as more reliable approach for aerosol correction.
This incongruent result may be related to the uncertain reliability of AERONET SSA level 1.5 of our data. Mainly due to the difficulty to determine absorbing aerosols in 20 many sites a detailed analysis in our area of study has been performed, based on the available information of AOD-alpha parameters, more frequently available than AI and SSA. The analysis allows the selection of some defined aerosol groups, as polluted or desert dust types. However, some inconsistence appeared. Polluted aerosols based on AOD-alpha values appear as non-absorbent according to AI values or moderate 25 absorbing according to SSA values and also a few percent of data shows low values of SSA (very absorbing) in contrast with AI values lower than 0.5 (non-absorbing). Only Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | desert dust data are in agreement for AOD-alpha, AI and SSA data. Most of the data of low AOD values appear as non-absorbing according to AI with moderate SSA values, then the applied corrections work. Proposed corrections for aerosol effect showed a relative high efficiency for reducing the OMI bias depending significantly on the slope of OMI/Brewer ratio over the absorp-5 tion aerosol optical depth. Our results reflect the level of accuracy that may be reached in this type of comparison at present which may be considered as satisfactory. Nevertheless, improvements must be accomplished to determine reliable absorbing aerosol properties, which appears as the limiting factor to improve OMI bias.   Table 2. Results of linear regression analysis between OMI and Brewer UV products without correction and the correction for the two approaches: SSA variable and SSA constant, and the statistical parameters of relative differences. The parameters are the following: the slope of the regression, the standard error (SE) of the slope, the Y intercept, the SE of the Y intercept, the correlation coefficients (R 2 ), and the root mean square errors (RMSE).           Fig. 11. Relative difference between OMI for UVER data as function of aerosol optical depth without correction when the two corrections are applied (a) for the first correction approach with SSA constant; (b) for the second correction approach with SSA variable.