Ozone Monitoring Instrument spectral UV irradiance products: comparison with ground based measurements at an urban environment

We have compared spectral ultraviolet overpass irradiances from the Ozone Monitoring Instruments (OMI) against ground-based Brewer measurements at Thessaloniki, Greece from September 2004 to December 2007. It is demonstrated that OMI overestimates UV irradiances by 30%, 17% and 13% for 305 nm, 324 nm, and 380 nm respec- 5 tively and 20% for erythemally weighted irradiance. The bias between OMI and Brewer increases with increasing aerosol absorption optical thickness. We present methodologies that can be applied for correcting this bias based on experimental results derived from the comparison period and also theoretical approaches using radiative transfer model calculations. All correction approaches minimize the bias and the standard de- 10 viation of the ratio OMI versus Brewer ratio. According to the results, the best correction approach suggests that the OMI UV product has to be multiplied by a correction factor C A ( λ ) are in the order of 0.8, 0.88 and 0.9 for 305 nm, 324 nm and 380 nm respectively. Limitations and possibilities for applying such methodologies in a global scale are also discussed.

tively and 20% for erythemally weighted irradiance. The bias between OMI and Brewer increases with increasing aerosol absorption optical thickness. We present methodologies that can be applied for correcting this bias based on experimental results derived from the comparison period and also theoretical approaches using radiative transfer model calculations. All correction approaches minimize the bias and the standard de-

Introduction
During the past decades a lot of effort has been put into composing global UV irradiance climatology datasets using ground based (GB) measurements from different locations (WMO, 2007). However, due to the limited availability of long term UV data series, combined with the fact that GB measurements cover a small fraction of the 20 Earth's surface, satellite sensors are widely used for estimating UV irradiance reaching the ground from ozone and reflectivity measurements, have gained particular attention. The development of satellite UV estimation techniques have been among the most important scientific issues within the UV community during the previous years (e.g. Krotkov et al., 2002). Following the many successful years that Total Ozone Map-One of the goals of OMI is the investigation of changes in global surface UV irradiance. Surface UV estimates based on satellite data have been used extensively in the last decade to establish global UV climatologies and to examine possible long-term 15 changes in surface UV (WMO, 2007 and references therein). The OMI surface UV algorithm consists of a calculation for clear sky cases extended with corrections in the case of cloudy pixels (or ones containing non-absorbing aerosols) (Tanskanen et al., 2006). In the real atmosphere, provided that cloud screening procedures are accurately taken into account, the accuracy of the UV product is limited by the imperfect 20 knowledge of aerosol properties and pollutants mainly in the boundary layer. Cloud effects have also a large impact in the UV retrieval from OMI and in addition they need to be accurately detected for reliable OMI aerosol retrieval. The small pixel-size of OMI, as compared with TOMS, is a major advantage for cloud screening. The validation of satellite-derived UV products against GB measurements is an essential task in 25 order to assess their accuracy. Several validation studies with heritage TOMS satellite instrument (e.g. Herman et al., 1999;Fioletov et al., 2002;Kazantzidis et al., 2006) revealed a positive bias in many locations, with the satellite derived UV being higher. It was also observed that satellite UV agrees much better with the GB measurements Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion at pristine sites (Lauder, in New Zealand (McKenzie et al., 2001), or Canadian West coast (Fioletov et al., 2002) than in more polluted European and North-American sites (Kazantzidis et al., 2006). Therefore, it was suggested that at least part of the bias could be attributed to the boundary layer aerosol absorption, not accounted for in the current satellite UV algorithm (Krotkov et al., 2002;Arola et al., 2005).

5
Lately, Tanskanen et al. (2008) focused on the validation of the widely used daily erythemally weighted irradiance. The satellite-derived erythemal daily doses were compared with those derived from the ground-based measurements at 17 sites and 18 instruments. The main objectives of this study were to discuss the applicability of the OMI data for surface UV monitoring and to establish a more accurate cloud correction 10 method for UV retrieval. One of their conclusions, confirmed also by Ialongo et al., 2008, is that the OMI UV aerosol absorption plays an important role in the UV radiation levels reaching the ground and for such cases OMI overestimates the calculated surface UV irradiance levels, especially at urban areas, by 20% to 40%. With regard to the spectral aspect, Wuttke et al. (2003) emphasized the importance of spectral compar- 15 isons within validation studies as a mean to provide possible hints about the sources of uncertainties.
In this study we focus on comparing OMI and GB UV spectral measurements at different wavelengths at the urban aerosol environment of Thessaloniki, Greece. The difference at each wavelength is quantified and categorized according to the presence 20 of clouds or aerosols in the area. Furthermore, we investigate possible off-line correction formulas in order to take into account aerosol absorption effects that influence the OMI UV.  Garane et al., 2006). In addition, direct irradiance spectra under cloud-free conditions and total ozone measurements have been used to derive aerosol optical thickness (AOT) (Kazadzis et al., 2007). The monitoring program includes a suite of other radiometric measurements and observations that are used for the characterization of the atmospheric conditions during the measurements. These ancillary measurements in-10 clude: solar irradiance at 302, 312, 320, 340 and 380 nm and erythemal irradiance from two NILU-UV multi-filter radiometers, erythemal irradiance from a YES UVB-1 radiometer, shortwave solar irradiance from a Kipp & Zonnen CM 21 pyranometer and sky images sampled every 5 min. These data sets have been used to compare irradiance measurements at the surface with the overpass irradiance data calculated from 15 OMI measurements. The monitoring station is located at the center of the city of Thessaloniki which is the second largest city in Greece, with a population of 1.2 million, and is bounded to the north by the Balkan countries and to the south by the Aegean Sea. Thessaloniki is affected by both anthropogenic and natural emission sources (Kazadzis et al., 2007). 20 Local emissions and regional pollution transport contribute to the poor air quality of the city which is reflected mostly in particulate matter (PM) concentrations. The 24 h limit values for PM 10 (50 µgr/m 3 ) are exceeded for more than half of the days during a year in the city centre's environmental monitoring sites. The amount of aerosols in this location is quite high, with a annual mean AOT at 340 nm equal to 0.45 and monthly values 25 ranging between 0.3 and 0.7, for winter and summer months, respectively (Kazadzis et al., 2007).
Comparisons of GB data with OMI products were conducted for the spectral irra-17471 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion diance at 305, 324, 380 nm and for the CIE weighted dose rate using GB averaged irradiance measurements performed within ±15 min from the OMI overpass time. It is well known that satellite derived surface irradiance is greatly affected by clouds and under such conditions comparisons with GB measurements at a single point is always challenging. For this reason, the comparisons were separated in cloud-free and all-5 skies data sets. Cloud free cases have been determined using the classification of the ground based measurements using pyranometer data (Vassaras et al., 2001), in combination with meteorological hourly sky observations. Brewer spectroradiometer UV data were corrected for wavelength shifts using the SHICRIVM algorithm (Slaper et al., 1995). The OMI derived spectral irradiance data 10 are reported at the same spectral resolution with the GB measurements (0.55 nm full width at half maximum). As the Brewer spectra are measured up to 365 nm, the irradiance data at 380 nm was constructed from each measured spectrum using the SHICRIVM algorithm. The time series of the OMI -GB irradiance were analyzed and the possible differences that could depend on various parameters such as: solar zenith 15 angle, ozone, time, presence of clouds and aerosols, were investigated.
The AOT information was taken into consideration in the following analysis only for clear sky days and only when at least one GB measurement was available within ±30 min from OMI overpass time. The associated single scattering albedo (SSA) was retrieved at 340 nm using the methodology described in Bais et al., 2005. Model cal-20 culations were performed using the Libradtran (Mayer and Kylling, 2005) package, in order to construct look up tables (LUT) of the ratio of UV irradiances, under aerosol free and actual aerosol conditions.

OMI UV retrieval methodology
The OMI surface UV algorithm is an extension of the TOMS UV algorithm developed 25 at NASA Goddard Space Flight Center (GSFC) (Eck et al., 1995;Krotkov et al., 1998;Herman et al., 1999;Krotkov et al., 2001;Tanskanen et al., 2006). The OMI surface UV algorithm is used for offline production of the global surface UV data using as input 17472 Introduction

Tables Figures
Back Close

Printer-friendly Version
Interactive Discussion the OMI TOMS total column ozone (Bhartia and Wellemeyer, 2002) and reflectance at non-absorbing wavelength (360 nm). The common approach involves estimation of the clear-sky surface irradiance E clear which is adjusted to actual surface irradiance by using a satellite derived cloud/aerosol transmittance factor C T : The a-priori surface albedo information required for modeling of the surface UV irradiance is not point-wise surface albedo but rather a regional quantity, referred to as effective albedo, and describes the overall effect of the surface albedo of the surrounding area on the surface UV irradiance. The radiative transfer model involves a climatological set of latitude-dependent ozone and temperature profiles. Krotkov et presented two alternative cloud correction methods: (1) based on modified Lambertian Equivalent Reflectivity (LER) model (Eck et al., 1995) and (1) based on homogeneous plane-parallel C1 cloud layer. LER is determined by solving a simple form of the radiative transfer equation that assumes a pure Rayleigh scattering atmosphere bounded by an isotropically scattering Lambertian surface (e.g., Bhartia et al., 1993) at terrain 20 pressure, and C T is estimated from equation: where R s is surface albedo. Absorbing aerosols (e.g., organic carbon, smoke, and dust) or trace gases (e.g., NO 2 , SO 2 ) are known to lead to systematic overestimation of the surface UV irradiance 25 (Krotkov et al., 1998;Arola et al., 2005;Chubarova, 2004). The current OMI surface UV algorithm assumes no aerosol absorption. In this work we introduce an additional 17473 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion aerosol absorption correction factor C A that can be applied to post-correct operational OMI UV data as follows: The C A is defined to be equal unity in case of non-absorbing aerosols (sulfate, sea salt) and less than unity in case of absorpbing aerosols (i.e. dust, carbonaceous, pollu-5 tion). In this paper we provide and compare several algorithms for estimating C A , which can be later applied for a post correction of the operational OMI UV data according to Eq. (3) to account for aerosol absorption effects. 10 The comparison of the Brewer spectroradiometer GB measurements and OMI standard UV product (E cloud ) was assessed by studying the OMI/GB ratio (RT). The histograms of RT are plotted for each of the three wavelengths and the CIE dose rate separately. The distributions were analyzed by calculating the median value of RT (m). Additionally we have determined the percentage of the satellite-derived data that agrees within ±10 15 and ±20% with the reference data (denoted as W 10 and W 20 , respectively). Cloudless cases statistics are also presented. In addition, and only for cloudless cases, GB measurements are compared with E clear to assess the calculated C T that for this cases includes only aerosol scattering correction.

Validation statistics
The validation statistics are presented in Table 1. Together with the median and the 20 W 10 and W 20 the coefficient of determination (R 2 ) is presented.
Results of Fig. 1 and Table 1 show an overall overestimation of the Ecloud compared with the GB measurements. The bias is larger for shorter wavelengths. The bias slightly increases (10%-15%) for all sky (cloudless or cloudy) conditions compared to the clear sky cases (Table 1). Also, the W 10 and W 20 percentages are dropped by ACPD 8,2008 OMI UV validation at an urban environment S. Kazadzis et al.  (1), it has to be noted that differences among E clear and 5 Brewer GB on cloudless conditions are in the order of 2-3% which are the levels of the correction that is included in the OMI UV algorithm for non-absorbing aerosol effects. In order to evaluate additional correction methodologies for absorbing aerosols we have been using the above dataset and measured results for the area of Thessaloniki.

UV attenuation due to aerosols
The Thessaloniki area is characterized by strongly aerosol load, with particularly absorbing aerosols and increased air pollution Koukouli et al., 2006;Amiridis et al., 2005). Thus it can easily be argued that the degree of overestimation of the Thessaloniki OMI surface UV irradiance under cloudless conditions is mainly due 15 to the aerosol absorbing layers not taken into account in the OMI UV algorithm.
Three year monthly mean AOT and total ozone climatological data from Thessaloniki have been used as inputs to the Libradtran radiative transfer model to calculate the attenuation of UV irradiance at 324 nm due to aerosols (Fig. 2). The SSA used for these calculations was set to 0.90 at 340 nm. Figure 2 also

Investigation of possible post-correction methods
Based in the above results, in this section we investigate various methodologies that could probably be used for correcting this aerosol absorption effect on the OMI retrieved 5 UV products and to calculate C A . For this purpose the Aerosol Absorption Optical Thickness (Ta) as is defined from the following formula, was used: The methodologies shown below are based on the dependence of RT to AAOT and can be: (a) experimental, using the results of GB and OMI measurements and (b) 10 theoretical, using radiative transfer model calculations.
The experimental method is based on the investigation of a similar TOMS to GB bias presented in Arola et al. (2005) which here is performed by comparing GB and OMI data under clear sky conditions. In Fig. 3 the ratio OMI versus GB measurements at 324 nm under clear sky conditions is shown against AAOT. 15 The AOT is measured within ±30 min from the OMI overpass at 324 nm and the SSA is derived from Brewer irradiance measurements based on the methodology by Bais et al. (2005). The calculated slope was found equal with 2.1 per unit of AAOT and with an intercept of 1.07. Using this slope we investigated if such a correction could improve the observed bias showed in Table 1. In addition, results presented by Krotkov 20 et al. (2005), that calculated RT through AOT, and SSA measurements at Washington area for TOMS satellite, proposed a similar correction using the formula: The method presented above is solar zenith angle independent. However, the effects of an aerosol absorbing layer will lead to higher UV attenuation at higher solar zenith Interactive Discussion angles due to the increased optical path of the solar photons through this layer. For that purpose we tried to eliminate the solar zenith angle effect on aerosol absorption following an idea of Garcia et al. (2006). We used slant absorption optical thickness (AAOTS) instead of AAOT, which is defined as: where sza is the solar zenith angle. Using the same dataset presented in Fig. 3 we calculated the RT dependence on AAOTS, shown in Fig. 4. This method provides a new slope of 1.6 per unit of aerosol slant column absorption optical thickness and an intercept of 1.06. The solar zenith angle impact on the effect of aerosol in the UV irradiance is crucial especially for locations with annual AOT and ). In addition, as the OMI overpass time is almost constant throughout the year, wintertime measurements are performed at higher solar zenith angles than the ones in summertime. All the above are not taken into consideration in the first method and could lead to comparable UV attenuation due to aerosols in winter and in summer despite the fact that AOT is higher in summertime. 20 Another approach that was investigated was to calculate C A with the use of model calculations. The absorbing aerosol layers that are "invisible" for the OMI instrument lead to the underestimation of the attenuation of UV irradiance reaching the ground. With the help of the Libradtran radiative transfer model we constructed look up tables (LUT) quantifying aerosol scattering and absorption due to an aerosol layer. An ex-25 ample is shown in Figs. 5a and 5b: Lines in Fig. 5a  and SSA for a constant solar zenith angle of 40 • and total column ozone of 340 DU. As expected the higher the AOT and the lower the SSA, lead to higher correction factors calculated. For a certain case of SSA (0.88 at 340 nm) and AOT of 0.5 at 340 nm the correction factor shown in Fig. 5b is ∼17%. While ∼6.5% of this is due to aerosol scattering (that is already taken into account on the OMI correction through C T ) and 5 ∼10.5% due to aerosol absorption, for UV irradiance at 324 nm. All the above methods include the premise that various parameters like AOT and SSA in the UV (preferably) or in the visible (through AERONET data) have to be known in order to be used for calculating C A on a global scale. We have used Thessaloniki data to investigate how the C A statistics derived from the theoretical method are altered when using a constant (climatological) AOT and SSA. For this purpose we have used synchronous AOT Brewer measurements at 340 nm and a mean SSA (0.91) at 440 nm for the 2004-2007period (Kazadzis et al., 2007, derived from a CIMEL sunphotometer. Through the calculated LUT's we have retrieved C A using cloud-free days only. The results of the statistics for all of the above correction methods are provided 15 in the following Table 2: The results presented in Table 2 are the OMI-GB ratios after applying each method's calculated C A 's. All methods presented here lead to an improvement of the OMI-GB bias. Especially method #3 (shown in this table), that is described by formula (5), provides the best correction approach. The use of either of the aforementioned methods 20 by any ground-based measuring station depends on the availability of independent aerosol optical properties data (AOT, SSA).
We have tried to investigate possible dependence of solar zenith angle, aerosol optical thickness, absorption optical thickness, ozone or time to the original and correction applied datasets. No systematic dependence of C A to the above parameters was 25 found. As an example, the results of the application of the above methods to the time series of RT measurements are shown in the following Fig. 6.
Using the C A 's retrieved by method #3 (on Table 2) we have calculated corrected OMI UV irradiances for 305 nm, 324 nm and 380 nm which are presented in Fig. 7 Confirming further the results of Table 2, the coefficients of determination calculated after applying the corrections are much improved. It has to be mentioned that the number of the original cases shown in Fig. 7 is 267 (cloudless cases over 3.5 years period), while the corrected ones are only 201 due to the absence of coincident aerosol optical thickness measurements within one hour from the OMI overpass.

Discussion and conclusions
Three and a half years of the operational OMI UV data and Brewer UV irradiance measurements were compared for the area of Thessaloniki, Greece. Lergest positive OMI biases were found at 305 nm that reach 32% and 27% for cloudy and cloudless measurements. Smaller biases were found for UVA wavelengths: 20% and 16% at 10 324 nm and 16% and 11% at 380 nm, respectively. The biases were attributed to the absorption of aerosols not taken into account from the OMI algorithm.
Various methods have been tested in order to investigate the possibility of an OMI UV data post correction for absorbing aerosols. All methods presented here lead to improved OMI-GB comparisons. Using AOT measurements at the exact wavelengths 15 analyzed and SSA measurements at 340 nm the correction methodologies were compared. Results showed that for 305 nm the averaged correction C A is compensating for most of the observed overestimation while for 324 nm and for 380 nm the correction methods eliminate most of the observed biases.
The post correction factors, C A calculated for Thessaloniki area using formula 5 are 20 in the order of 0.8, 0.88 and 0.9 for 305 nm, 324 nm and 380 nm, respectively. Such values can be used for urban areas with similar or measured aerosol optical properties.
Using the slopes of RT as a function of AAOT, measured at Thessaloniki, there remains a 9%, 6% and 4% positive OMI bias not related to aerosol absorption. A possible explanation for this could be the fact that in cloudless cases, absorbing aerosols attenuate 25 also the reflected irradiance from the ground. That could lead to a small underestimation of LER resulting to a small C T overestimation (see Eq. 2). The final estimated

ACPD Introduction
Conclusions References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion surface irradiance from OMI will be overestimated especially at low SSA cases. On the contrary, using formula (5) which introduces a larger slope than the one measured, in a way, this effect is eliminated. Spectral differences of methods 1 and 2 presented in Table 2 can be also a result of spectral dependence of the SSA. Comparing total column ozone values retrieved from OMI and GB measurements we 5 found a small OMI ozone under-estimation in the order of 1.2% (mean), which can be also have a (smal) effect to all OMI-GB comparisons, especially at 305 nm. Model calculations using measurements of a CIMEL sun-photometer were introduced because of the availability of such measurements (especially SSA) from the specific instruments worldwide. The method retrieves larger C A and the corrected RT's 10 still show OMI overestimation especially for lower wavelengths. One possible explanation for the above results could be the spectral dependence of SSA. Previous works  indicated that SSA could be less in UVB than in the visible wavelength range especially at the urban area under investigation. This SSA effect becomes more complicated when considering seasonal variations of absorbing aerosol proper-15 ties. As the mean SSA at 440 nm for the area is 0.91 using a value of 0.88 for correction at 380 nm, leads to a few per cent differences among the corrected OMI irradiances and the GB ones. However, an SSA value that could be used for achieving similar results for the 305nm irradiance would be approximately 0.84. Measurements of SSA in the UVB are still uncertain especially when AOT is low and further investigation on the 20 absorbing aerosol and gaseous (NO 2 , SO 2 ) properties at these wavelengths is still an intriguing scientific future issue.
The standard deviation for all cloudless OMI vs Brewer ratios were found to be reasonable, taking into account possible OMI sub-pixel variability of AOT and thus UV irradiances. Results of OMI sub-pixel UV variability showed that for pixels that include 25 urban areas UV variability can be as high as 20% even for cloudless conditions. While a similar campaign in a non urban environment (Weihs et al., 2008) showed that for cloudy conditions UV variability can reach 50%. The above results show the limitations when comparing satellite with single point GB measurements. For the case of cloudy conditions the OMI-Brewer ratios showed approximately 5% higher values than the cloud-free ones and in addition higher standard deviation. For the higher values the explanation is the high UV irradiance variability under cloudy conditions and the absence of information on the amount of the direct sun attenuation for each individual measurement. An additional reason is the fact that OMI overpasses 5 and GB spectral measurements are not exactly synchronous, which is important issue comparing UV measurements under a variable radiation field. The absolute level differences and their spectral dependence indicate that similar C A 's have to be applied in cloudy cases too. However, the UV attenuation due to aerosol absorption under cloudy conditions is more difficult to be studied theoretically, as more information about the 10 clouds and the aerosol layer height are necessary. In addition, aerosol absorption is difficult to be quantified as sun-photometric measurements fail under cloudy conditions. In this case, the calculated C A 's can be used, or other possible correction procedures have to be considered.

ACPD
Post processing algorithms for OMI-UV products have to be employed when using 15 OMI data for regions with high aerosol load. Simple parameters for public use (e.g. UV Index) can be overestimated by more than 20% if using OMI overpass data for urban areas such as Thessaloniki. The use of climatological mean values of AOT and absorption (SSA) aerosol quantities in a global scale could be used based in one of the methodologies presented in this work. Finally, when it is possible, the spectral depen-20 dence of AOT and SSA has to be taken into account in order to include a correction for the spectral dependence of UV attenuation reaching the ground, in the presence of an aerosol layer.