Measuring aerosol UV absorption by combining use of shadowband and almucantar techniques N.A. Krotkov, Goddard Earth Sciences and Technology Center /UMBC and NASA/GSFC P.K.Bhartia, J. Herman, NASA/GSFC, Jim Slusser , Gwen Scott, USDA UVB Monitoring network G. Labow, A.Vasilkov, SSAI T. Eck, O. Dubovik, GEST/UMBC and B. Holben, NAS A/GSFC

Why is aerosol UV absorption important ? Aerosol effects on UV trends may enhance reduce, or reverse effects of stratospheric O3 change TOMS overestimation of surface UV irradiance 22 21 + 10%-20% 23 3) Aerosol effects on photochemical smog production: aerosol scattering increases photolysis rates; while aerosol absorption decreases it: Change in BL ozone mixing ratios as a result of direct aerosol forcing: +20ppb ( =0.95) -24ppb ( =0.75)

Why is aerosol UV absorption important ? EP TOMS (1996-) AURA/OMI (2004 - ) 24 New TOMS/OMI aerosol UV absorption product needs validation

Possibility exist to derive column aerosol absorption from the ground: (1) Direct sun radiation ~ aerosol extinction () (2) Diffuse sky radiation ~ aerosol scattering (**) Combining (1) and (2) measurements allows to derive aerosol single scattering albedo:  and absorption optical thickness: *(1-)

Practical implementation: (1) Diffuse-To-Direct Irradiance technique First proposed by B. Herman, R.S.Browing and J. J. De Luisi [JAS,1975] and implemented by J.J. De Luisi [1976] and M. King [JAS, 1979] in the VIS was recently used by T. Eck et al [JGR 2003] in VIS and by Wenny et al [1998], Petters et al [JGR, 2003], Wetzel et al [2003] and C.Goering, et al. [2004] in UV All recent UV measurements were conducted with UV-MFRSR (YES) [L. Harrison and J. Michalsky ]

USDA UV-B Monitoring and Research Program operates US network of UV MultiFilter Rotating Shadowband Radiometers (UV-MFRSR) http://uvb.nrel.colostate.edu UVMFRSR was continuously operated at NASA/GSFC since October 2002 3 min measurements of total and diffuse irradiance measurements at 300, 305, 311, 117, 325, 332, 368nm

Daily Vo Calibration Transfer Mauna Loa solar calibration Daily Vo Calibration Transfer Diffuse/Total fraction and  Single scattering albedo,  RT model fitting Sphere radiometric calibration A-priori information The aerosol extinction optical thickness, ext, was measured by the CIMEL direct-sun technique in the visible and at two UV wavelengths 340 and 380 nm. These results were used for UV-MFRSR daily on-site calibration and measurements of ext at 368nm . The ext(368) measurements were used as input to a radiative transfer model along with AERONET retrievals of the column-integrated particle size distribution (PSD) to infer an effective imaginary part of the UV aerosol refractive index, k. This was done by fitting MFRSR measured voltage ratios with the radiative transfer model. Inferred values k368 allow calculation of the single scattering albedo, 368, and comparisons with AERONET 440 retrievals.  extinction by AERONET at 340nm –500nm AERONET also measures sky radiances enabling retrieval of size distribution and effective spectral refractive index at 440nm - http://aeronet.gsfc.nasa.gov

Error estimate  ~ abs / ~ 0.01/   ~ 0.02 (~0.5 ) Error in absorption optical thickness:  abs (368nm) ~0.01 - 0.02 ( limited by the measured accuracy of total voltage and calibration, V0) Error in single scattering albedo :  ~ abs / ~ 0.01/   ~ 0.02 (~0.5 ) Error due to uncertainty in size distribution and real refractive index becomes comparable to the measured uncertainties only for large aerosol loadings (ext>0.5)

Siberian smoke plume on June 2, 2003 AERONET SSA UV-MFRSR AOT Figure shows  retrievals by both instruments on June 2 2003, when a long-range smoke plume was moving over GSFC location. The passage of the plume is evident from enhanced extinction optical thickness, 368, measured by both instruments (shown on the right axis in figure 2). Visually, horizontal visibility remained high on this day with clear sight of horizon; however, the sky color was unusually white. According to 3-min MFRSR data, the most absorbing part of the smoke plume (368 ~0.88-0.9) was recorded in the morning (<14UT) with less absorbing 368 ~0.93 for the rest of the day. Back trajectory analysis and satellite data suggested that the smoke plume was originated from fires in Siberia near lake Baikal. Physical-chemical processes during long-range transport of smoke can explain this relatively low absorption. Boreal forest smoke typically does not have low  due to significant particle production from smoldering of woody fuels, which yields relatively small black carbon percentages. Also smoke particles tend to become less absorbing with age as the particle size increases due to coagulation during transport36. The Angstrom exponent was high and stable during the day (440/870=1.73-1.88), suggesting predominantly fine-mode particles. However, the Angstrom exponent was smaller in UV (380/440=0.73-0.82) compared to the visible wavelengths. This suggests substantial curvature of the ln() vs ln() dependence (’=1.7-1.8)24,29. The cause of large  discrepancy in the morning (~11.5UT) remains unknown

Diurnal 368 dependence on June 24, 2003 AERONET SSA UV-MFRSR AOT Slide shows  comparisons on June 24 2003, which was typical for a summer local ozone pollution episode. A high-pressure system over the Mid-Atlantic region for this week prevented air exchange; therefore tropospheric ozone pollution was building up as a result of local pollution (mostly traffic) and high solar insulation3 (air quality public warning was “Code orange” on June 24). The conditions were mostly cloud-free for this day. In the morning aerosol absorption was higher in UV, but differences were not significant. Aerosol extinction decreased during the day, while absorption increased slightly, but more rapidly in the visible. In the afternoon both retrievals show excellent agreement (368=0.91-0.92, k368=0.011-0.014, 440=0.91-0.92, k440=0.01-0.012). AERONET real part of refractive index at 440nm changed between 1.39 and 1.59. The Angstrom exponent was much higher than for smoke event on June 2, especially in UV (Table 2, day 175) due to significantly smaller radius and broader  of the fine mode on June 24.  

Diurnal 368 changes on August 25, 2003 AERONET SSA AOT UV-MFRSR We selected  to demonstrate retrievals on August 25, 2003 (Figure 4) because of strong daily variation in 368 with unusually low values (368 ~0.85) in the middle of the day. This case highlights the importance of measuring the complete diurnal cycle of summertime aerosol absorption, not just morning and afternoon periods.

Comparison statistics (1): Extinction  (all clear sky cases ~10,000)  368 < 0.02 (daily rms differences) for all clear sky days   < 0.01 (daily rms differences) for  < 0.4 Daily aerosol extinction optical thickness rms differences between UV-MFRSR and AERONET CIMEL measurements at 368nm. 325nm Inferred values of the effective UV imaginary refractive index were used for comparisons of aerosol single scattering albedo, 368, at 368nm with AERONET retrievals at 440nm, 440, using the Dubovik and King algorithm25. The measured differences in absorption between 368nm and 440nm might suggest the presence of selectively UV absorbing aerosols5 or interference from gases other than ozone34. In our opinion, the 368 results we reported do not allow confident explanation of the causes of apparent larger absorption at 368nm compared to 440nm. Continuing co-located measurements at GFSC location is important to improve the comparison statistics, but conducting these measurements at different sites with varying background aerosol conditions is also desirable. The differences in k were consistent with lower  values in UV (368=0.89-0.92 compare to 440=0.93-0.95). This suggests that  spectral dependence in the visible (lower  at longer wavelengths)27 flattens out and even reverses in UV. However, it is emphasized that, except for SZA>70o , the  retrieved at 368nm and 440nm are within the range of overlap of retrieval uncertainties It should be mentioned for AERONET wavelengths, that the  increases with decreasing  for fine mode smoke or pollution aerosol27. Therefore, the extrapolated differences in 368 (predicted by AERONET) and 368 retrieved by UV-MFRSR may be slightly greater than direct comparisons of 440 to 368. Therefore, it is important in the future to conduct similar measurement in locations affected by desert dust or more polluted sites, where UV aerosol absorption is expected to be higher 368nm

Comparison statistics (2): single scattering albedo (65 matchup cases in summer 2003)  case average ( 65 matchup cases in Summer 2003) <368>=0.93 +/-0.02 (1) at 368nm <440> =0.95 +/-0.02 (1) at 440nm  mean difference comparable to retrieval uncertainty <440 - 368> ~0.02, rms difference: ~0.016 Inferred values of the effective UV imaginary refractive index were used for comparisons of aerosol single scattering albedo, 368, at 368nm with AERONET retrievals at 440nm, 440, using the Dubovik and King algorithm25. The measured differences in absorption between 368nm and 440nm might suggest the presence of selectively UV absorbing aerosols5 or interference from gases other than ozone34. In our opinion, the 368 results we reported do not allow confident explanation of the causes of apparent larger absorption at 368nm compared to 440nm. Continuing co-located measurements at GFSC location is important to improve the comparison statistics, but conducting these measurements at different sites with varying background aerosol conditions is also desirable. The differences in k were consistent with lower  values in UV (368=0.89-0.92 compare to 440=0.93-0.95). This suggests that  spectral dependence in the visible (lower  at longer wavelengths)27 flattens out and even reverses in UV. However, it is emphasized that, except for SZA>70o , the  retrieved at 368nm and 440nm are within the range of overlap of retrieval uncertainties It should be mentioned for AERONET wavelengths, that the  increases with decreasing  for fine mode smoke or pollution aerosol27. Therefore, the extrapolated differences in 368 (predicted by AERONET) and 368 retrieved by UV-MFRSR may be slightly greater than direct comparisons of 440 to 368. Therefore, it is important in the future to conduct similar measurement in locations affected by desert dust or more polluted sites, where UV aerosol absorption is expected to be higher

Comparison statistics (3): Imaginary part of refractive index (65 matchup cases in summer 2003) Higher values for imaginary refractive index, k in UV: <k368> ~0.01, k368~0.004 at 368nm <k440> ~0.006, k440~0.003 at 440nm Inferred values of the effective UV imaginary refractive index were used for comparisons of aerosol single scattering albedo, 368, at 368nm with AERONET retrievals at 440nm, 440, using the Dubovik and King algorithm25. The measured differences in absorption between 368nm and 440nm might suggest the presence of selectively UV absorbing aerosols5 or interference from gases other than ozone34. In our opinion, the 368 results we reported do not allow confident explanation of the causes of apparent larger absorption at 368nm compared to 440nm. Continuing co-located measurements at GFSC location is important to improve the comparison statistics, but conducting these measurements at different sites with varying background aerosol conditions is also desirable. The differences in k were consistent with lower  values in UV (368=0.89-0.92 compare to 440=0.93-0.95). This suggests that  spectral dependence in the visible (lower  at longer wavelengths)27 flattens out and even reverses in UV. However, it is emphasized that, except for SZA>70o , the  retrieved at 368nm and 440nm are within the range of overlap of retrieval uncertainties It should be mentioned for AERONET wavelengths, that the  increases with decreasing  for fine mode smoke or pollution aerosol27. Therefore, the extrapolated differences in 368 (predicted by AERONET) and 368 retrieved by UV-MFRSR may be slightly greater than direct comparisons of 440 to 368. Therefore, it is important in the future to conduct similar measurement in locations affected by desert dust or more polluted sites, where UV aerosol absorption is expected to be higher

W results: spectral dependence On top of the annual cycle, occasional transient phenomena (long-range transport of biomass burning smoke and desert dust storms) can be clearly detected. One clear example is the passage of an aged smoke plume from Siberian forest fires over GSFC on June 2, 2003, characterized by an unusually large abs(368)~0.085. Although occasional dust plumes had been reported at GSFC (April 2001 Asian dust plume), no dust events occurred during reported time period. 368nm

368nm – 332nm

368nm –332nm- 325nm The real part of refractive index at 440nm increased from 1.39 to 1.5 during smoke passage and decreased later to 1.46. The imaginary part of refractive index was higher in UV than in the visible ( k368=0.014 - 0.02, k440=0.007 – 0.013 ). The difference was larger than specified uncertainty for AERONET k retrievals (+/-0.003)27 for all cases except one retrieval. These differences in k were consistent with lower  values in UV (368=0.89-0.92 compare to 440=0.93-0.95). This suggests that  spectral dependence in the visible (lower  at longer wavelengths)27 flattens out and even reverses in UV. However, it is emphasized that, except for SZA>70o , the  retrieved at 368nm and 440nm are within the range of overlap of retrieval uncertainties

W spectral dependence The spectral ratio does not remain constant , but changes during the day. So, potentially some useful information could be retrieved from the spectral dependence of W

368nm August 25, 2003

368nm – 332nm

368nm – 332nm –325nm Spectral dependence is flatter on Aug 25 than on June2 The cause for low W values at noon time remains unknown.

W spectral dependence The spectral ratio does not remain constant , but changes during the day. In this case W spectral difference exists only between 368 and 332nm, while 332 and 325 channels looks the same. There is distinct change round noon time. (the reason is unknown). So, potentially some useful information could be retrieved from the spectral dependence of W

Results: spectral dependence in UV-VIS ? ? SINGLE SCATTERING ALBEDO UV-MFRSR VIS- CIMEL It should be mentioned for AERONET wavelengths, that the  increases with decreasing  for fine mode smoke or pollution aerosol27. Therefore, the extrapolated differences in 368 (predicted by AERONET) and 368 retrieved by UV-MFRSR may be slightly greater than direct comparisons of 440 to 368. Therefore, it is important in the future to conduct similar measurement in locations affected by desert dust or more polluted sites, where UV aerosol absorption is expected to be higher UV MFRSR VISIBLE AERONET Wavelength

Aerosol absorption optical thickness: Seasonal Dependence The abs values show a pronounced seasonal dependence of ext with maximum values abs ~0.05 at 368nm (~0.07 at 325nm) occurring in summer hazy conditions and <0.02 in winter-fall seasons, when aerosol loadings are small. The maximum abs typically occurs in summer due to combination of regional and local pollution sources with hot and humid weather conditions (summer haze). The weakly absorbing haze (368>0.9) is often associated with high levels of tropospheric ozone (ozone smog episodes)3. These summer haze conditions are responsible for summer high abs values (at 368nm ~0.06-0.07). Even on relatively clear summer days abs is larger than ~0.02.

W results: Seasonal Dependence No clear W seasonal dependence The maximum abs typically occurs in summer due to combination of regional and local pollution sources with hot and humid weather conditions (summer haze). The weakly absorbing haze (368>0.9) is often associated with high levels of tropospheric ozone (ozone smog episodes)3. These summer haze conditions are responsible for summer high abs values (at 368nm ~0.06-0.07). Even on relatively clear summer days abs is larger than ~0.02.

W results: AOT dependence The decrease of single scattering albedo with optical thickness suggests that the type of aerosol changes between summer and winter conditions. The decrease of single scattering albedo with optical thickness suggests that the type of aerosol changes between summer and winter conditions. It is well known that aerosol in the mid-Atlantic region in summer is strongly hydroscopic35, therefore particle growth by swelling at high relative humidity may be partly the reason for reduced absorption in summer. Indeed, annual cycle of 368 is the same as ext annual cycle: with maximum in summer and minimum in winter. Obviously continuation of uninterrupted UV-MFRSR measurements at GSFC site is important to increase statistical significance of reported data.

Summary Deriving complete diurnal cycle of aerosol absorption The shadowband method is complementary to the AERONET almucantar retrieval of , because retrievals are more reliable at low solar zenith angles; absolute sky radiances calibration is not required; The variability in aerosol size distribution and real refractive index becomes comparable to the measured uncertainties only for large aerosol loadings (ext>0.5) Combined use of both methods allows: Deriving complete diurnal cycle of aerosol absorption Considering days with low aerosol loadings, thus obtaining complete seasonal cycle of aerosol absorption Extending spectral dependence of single scattering albedo into UV wavelengths 1)       We demonstrated that combined use of CIMEL sun and sky radiance measurements in the visible with UV-MFRSR total and diffuse irradiance measurements in UV provide a important advantages for remote measurements of column aerosol absorption across the UV-Visible spectral range. The shadowband method13,20-24,38-39 is complementary to the Dubovik and King25 almucantar retrieval of aerosol single scattering albedo, because the retrievals are more reliable at low solar zenith angles and absolute sky radiances calibration is not required. Therefore, combined use of both methods allows deriving the complete diurnal cycle of aerosol absorption. There are also advantages specific for the UV spectral region: (1) the surface albedo is much smaller in the UV than in the visible; (2) ext is larger (for the same aerosol mass) than in the visible spectral range; (3) shorter wavelength UVB channels of UV-MFRSR are being used for measuring total ozone, which in turn is required for correction of aerosol measurements.

Future work Continuing co-located measurements at GFSC location is important to improve the comparison statistics; Extending UV-MFRSR spectral coverage to 440nm and CIMEL almucantar retrievals to 340 and 380nm to allow for spectral overlap between 2 types of aerosol absorption measurements; Conducting simultaneous measurements at different sites with varying background aerosol conditions is desirable. 1)       We demonstrated that combined use of CIMEL sun and sky radiance measurements in the visible with UV-MFRSR total and diffuse irradiance measurements in UV provide a important advantages for remote measurements of column aerosol absorption across the UV-Visible spectral range. The shadowband method13,20-24,38-39 is complementary to the Dubovik and King25 almucantar retrieval of aerosol single scattering albedo, because the retrievals are more reliable at low solar zenith angles and absolute sky radiances calibration is not required. Therefore, combined use of both methods allows deriving the complete diurnal cycle of aerosol absorption. There are also advantages specific for the UV spectral region: (1) the surface albedo is much smaller in the UV than in the visible; (2) ext is larger (for the same aerosol mass) than in the visible spectral range; (3) shorter wavelength UVB channels of UV-MFRSR are being used for measuring total ozone, which in turn is required for correction of aerosol measurements.

Backup UV-MFRSR on-site calibration method is based on AERONET CIMEL measurements of aerosol extinction optical thickness (ext) by direct a sun technique and Langley Mauna Loa calibrations. The AERONET ext was interpolated or extrapolated to the UV-MFRSR wavelengths and measurement intervals and used as input to the spectral band model along with column ozone and surface pressure measurements to estimate zero air mass voltages, Vo in each UV-MFRSR channel. The method does not require stability of ext and allows independent Vo estimations for every individual 3-min UV-MFRSR measurement. Daily <Vo> estimates were obtained for cloud-free conditions and compared with the on-site Langley plot technique. On the clearest stable days both techniques agree within 1%. Uncertainties in ext measurements were estimated using co-located identical CIMEL instruments and considered as part of combined V0 error budget. Since both components are measured by the same diffuser/filter/detector combination, diffuse and total atmospheric transmittances are obtained directly from the voltage ratios: TD =VD/V0 and TT =VT/V0. Absence of diurnal V0 patterns suggests consistency between CIMEL and UV-MFRSR measurements Same calibration (V0) is applied to diffuse and total voltages

Practical implementation: (2) sun and sky radiance technique AERONET/CIMEL Sun photometers provide global extinction measurements by direct sun technique from 340nm to 1020nm AERONET network of CIMEL sun/sky photometers provide global direct sun measurements of tau in UV (340nm). The direct measurement of single scattering albedo is not possible by any technique without an inversion procedure. For example, AERONET infer  by inverting sky radiance measurements in almucantar in addition to tau measurements. Inversion procedures require a-priori information and certain assumptions to be made. - http://aeronet.gsfc.nasa.gov

> 440nm CIMEL sky radiance almucantar measurements/inversions: AERONET measures sky radiances enabling retrieval of column size distribution and effective spectral refractive index m=n() – ik() in VIS Measurements : t(l) and I(l,Q) l = 0.44, 0.67, 0.87, 1.02 mm 2o ≤ Q≤ 150o (up to 30 angles) AERONET also provide estimations of aerosol size distribution and refractive index at visible wavelengths. In principle this information can be used for tau and omega calculation at UV wavelengths using Mie theory. > 440nm measurements

SSA retrieval method: Fitting measured diffuse fraction with RT model by iterating on imaginary part of refractive index separately at each wavelength; Using MFRSR extinction optical thickness and AERONET inverted particle size distribution and real part of refractive index at 440nm as input to RT model Fixed surface albedo 0.02 and actual pressure scaling of Rayleigh optical thickness Measured Total ozone is used for calculation of ozone optical thickness Using Mie code to calculate single scattering albedo The aerosol extinction optical thickness, ext, was measured by the CIMEL direct-sun technique in the visible and at two UV wavelengths 340 and 380 nm. These results were used for UV-MFRSR daily on-site calibration and measurements of ext at 368nm . The ext(368) measurements were used as input to a radiative transfer model along with AERONET retrievals of the column-integrated particle size distribution (PSD) to infer an effective imaginary part of the UV aerosol refractive index, k. This was done by fitting MFRSR measured voltage ratios with the radiative transfer model. Inferred values k368 allow calculation of the single scattering albedo, 368, and comparisons with AERONET 440 retrievals.

SENSITIVITY OF UV-MFRSR MEASUREMENTS TO AEROSOL ABSORPTION The regression coefficients quantify TR sensitivity to aerosol parameters as function of solar zenith angle (Table 1). The expected accuracy of abs retrieval from MFRSR measurements is ~0.008-0.02 limited by the measured accuracy of total voltage (VT) and calibration (V0). The variability in aerosol size distribution and real refractive index becomes comparable to the measured uncertainties only for large aerosol loadings (ext>0.5). The measurement uncertainties (discussed in detail in the first paper38) and regression coefficients for high and low aerosol loadings are summarized in Table 1. We note that estimated retrieval uncertainties of abs and  for shadowband technique (Table 1) are comparable to almucantar technique25-27 for favorable conditions (large solar zenith angles, o >45o and high aerosol loadings ext(440)>0.4). However, an important advantage of the shadowband technique is that it remains sensitive to abs even at low solar zenith angles, when the almucantar technique is not sensitive to abs.25-27 On the other hand, cosine correction errors increase for shadowband measurements at high solar zenith angles (see discussion in the first part), while cosine errors are absent for the CIMEL. Thus, the two types of measurements are required for measuring complete diurnal cycle of aerosol absorption.   Relationship between Rayleigh normalized total transmittance, TR and abs at 368nm, assuming fixed ext=0.167 (red) and 0.2 (purple) and o=33o,70o. Linear regression model (1) is fitted to al data points assuming variability due to size distribution as random errors

Aerosol UV absorption experiment (2002-04): 1. UV-MFRSR calibration and performance at GSFC

- http://aeronet.gsfc.nasa.gov  absorption in UV Mauna Loa solar calibration Daily Vo Calibration Transfer Global measurements of direct sun aerosol extinction  by AERONET at 340nm and 380nm UV-MFRSR on-site calibration method is based on AERONET CIMEL measurements of aerosol extinction optical thickness (ext) by direct a sun technique and Langley Mauna Loa calibrations. The AERONET ext was interpolated or extrapolated to the UV-MFRSR wavelengths and measurement intervals and used as input to the spectral band model along with column ozone and surface pressure measurements to estimate zero air mass voltages, Vo in each UV-MFRSR channel. The method does not require stability of ext and allows independent Vo estimations for every individual 3-min UV-MFRSR measurement. Daily <Vo> estimates were obtained for cloud-free conditions and compared with the on-site Langley plot technique. On the clearest stable days both techniques agree within 1%. Uncertainties in ext measurements were estimated using co-located identical CIMEL instruments and considered as part of combined V0 error budget. Since both components are measured by the same diffuser/filter/detector combination, diffuse and total atmospheric transmittances are obtained directly from the voltage ratios: TD =VD/V0 and TT =VT/V0. Absence of diurnal V0 patterns suggests consistency between CIMEL and UV-MFRSR measurements Same calibration (V0) is applied to diffuse and total voltages USDA UV-B Monitoring and Research Program - http://aeronet.gsfc.nasa.gov http://uvb.nrel.colostate.edu

UV-MFRSR on-site calibration (V0) UV-MFRSR cosine corrected direct-normal voltage UV-MFRSR spectral band model Direct pressure measurements UV-MFRSR on-site calibration method is based on AERONET CIMEL measurements of aerosol extinction optical thickness (ext) by direct a sun technique and Langley Mauna Loa calibrations. The AERONET ext was interpolated or extrapolated to the UV-MFRSR wavelengths and measurement intervals and used as input to the spectral band model along with column ozone and surface pressure measurements to estimate zero air mass voltages, Vo in each UV-MFRSR channel. The method does not require stability of ext and allows independent Vo estimations for every individual 3-min UV-MFRSR measurement. Daily <Vo> estimates were obtained for cloud-free conditions and compared with the on-site Langley plot technique. On the clearest stable days both techniques agree within 1%. Uncertainties in ext measurements were estimated using co-located identical CIMEL instruments and considered as part of combined V0 error budget. Interpolated CIMEL a corrected for pressure Brewer/TOMS total ozone measurements Airmass factor

MFRSR: cosine corrected voltages We derive individual V0 by numerically integrating high resolution spectral transmittance TR() within each filter bandpass: UV-MFRSR angular response function, normalized to the ideal (cosine) angular response( fR=1 for ideal instrument). Instrument head #271, at 368nm measured by CUCF on 5 September 2002 before deployment at GSFC site. For each of the 7 channels, there are 2 sets of responses: one from the south to north scan (labeled SN), and one from the west to east scan (WE). Correction factor of the measured direct-normal voltage, fR, is interpolated according to solar zenith and azimuth angles at the time of actual measurement. The azimuth angle,  (0o-359o), is resolved into the four quadrants and normalized angular response factor for direct voltage, fR, is calculated as described by Harrisosn and Michalsky16.

UV-MFRSR spectral band model MFRSR spectral band model takes into account actual UV-MFRSR spectral response functions (SRF) as well as spectral variation of the solar flux and atmospheric extinction within each filter bandpass of the instrument. Our spectral band model (Figure 5, Table 2) takes into account actual UV-MFRSR spectral response functions (SRF) as well as spectral variation of the solar flux and atmospheric extinction within each filter bandpass of the instrument. We derive individual V0 by numerically integrating high resolution spectral transmittance TR() within each filter bandpass: In equation (2) the denominator represents absolute bandpass average solar flux, Etop [W/m2/nm] that would be measured by UV-MFRSR at the top of the atmosphere (shown is figure 5 and Table 2), while numerator represents attenuated solar flux measured at the surface Etop [W/m2/nm]. To evaluate the integrals one needs to know the extraterrestrial solar flux, E0, normalized spectral response function, F (min , max are channel cutoff wavelengths), atmospheric spectral extinction optical thickness and relative airmass factor, m. The UV-MFRSR normalized spectral response functions (SRF) for all channels are shown in figure 5 and the spectral band model is given in Table 2 for a nominal aerosol model. In practice, we use AERONET interpolated/extrapolated a to calculate TR() within each bandpass and numerically evaluate spectral integral in (2). We also calculate the radiatively equivalent wavelength in each channel, rad defined by equation (2).

Extrapolating AERONET AOT AERONET direct sun aerosol extinction optical thickness at 340nm, 380nm, 440nm, and 500nm normalized by (380). We extrapolate or interpolate AERONET ext measurements at 340nm, 380m, 440nm, and 500nm to the UV-MFRSR wavelengths using a least squares quadratic fit in log-log space (Figure 6)23. One could also use linear extrapolation in log-log space (dashed curve in Figure 6) assuming that the Angstrom parameter does not change. Although this assumption obviously does not hold for large wavelength spans (~100nm), both methods provide practically identical results in the UVA spectral region (320nm - 400nm). In our calibration technique, we use both methods and discard extrapolated values if they differ more than 0.01. At 368nm we interpolate rather than extrapolate and both methods typically provide practically identical results (within 0.003, figure 6). Therefore, we feel that spectral extrapolation is a negligible source of error in our calibration technique for longer UV-MFRSR channels. Extrapolation to longer UV-MFRSR channels using quadratic least squares fit of ln() versus ln() and linear extrapolation from 340nm and 380nm. The difference in ext between quadratic and linear interpolation methods is typically less than 0.005 at 368nm.

Daily <ln(V0)> AOT comparisons Standard deviation of Vo daily data, lnV0 , provides an overall measure of atmospheric and instrumental variability on a given day and is directly related to the uncertainty in derived individual ext values. On the clearest days at GSFC the scatter in Vo can be quite small (lnV0 ~0.006 for ext~0.1,figure 7, low panel), but the scatter increases on days with high turbidity (lnV0 ~0.03 for ext~0.6, figure 7, upper panel). In perfectly stable atmospheric conditions the Langley plot calibration method would provide an ultimate check on the measurements, spectral band model and <V0> calibrations17. Standard Langley technique regresses ln(Vn) versus m, so ln(Vo) is obtained as the zero airmass intercept of a linear regression model given by equation (3). This method does not require any a-priori knowledge of the atmospheric transmittance, beyond the stability requirement, which was difficult to meet at our site. Therefore, we tried to optimize standard Langley technique by adjusting the time interval used in regression to ensure maximum possible stability of ext during calibration. Figure 7 shows Langley technique Co results (with 1Co error bars) for different 2-hour calibration periods. We see that Langley technique produces close Vo results (within 1 error bars), when ext ~const. AOT comparisons

but the scatter increases on days with high turbidity (lnV0 ~0 but the scatter increases on days with high turbidity (lnV0 ~0.03 for ext~0.6,,). Calculating daily means of individual V0 values reduces the effect of random errors for transferring the AERONET calibration by the square root of the number of available measurements. Typically there are ~60 daily V0 estimates in winter compared to ~200 in summer, therefore, random error in estimating daily mean <V0> value (shown as a horizontal line in figure ) is reduced by a factor 8 to 20. The effect of random errors was further reduced by removing outlier measurements (with ln(V0) outside of +/-3lnV0 of the <ln(V0)>) and iteratively re-calculating <V0>. We found that removing less than 5% of outliers reduces lnV0 by half on both clear and turbid days (in figure lnV0 are shown for 3 iterations). AOT comparisons

Vo as diagnostic tool: diffuser cleaning after cleaning: +4% before cleaning Examining diurnal trends in V0 data provides insight into possible systematic calibration errors and yields a tool for checking consistency between AERONET-CIMEL ext and UV-MFRSR voltage measurements. Indeed, for perfect measurements, V0 should remain constant during the day regardless of any changes in atmospheric conditions, solar elevation, and azimuth. For example, any systematic residual errors in UV-MFRSR cosine or shadowing corrections will manifest themselves as systematic ln(V0) changes with solar zenith angle. Thus, a longer calibration period permits a better estimate of possible systematic errors.

Vo as diagnostic tool: incomplete shadowing Morning meas. OK Shadowing problem Examining diurnal trends in V0 data provides insight into possible systematic calibration errors and yields a tool for checking consistency between AERONET-CIMEL ext and UV-MFRSR voltage measurements. Indeed, for perfect measurements, V0 should remain constant during the day regardless of any changes in atmospheric conditions, solar elevation, and azimuth. For example, any systematic residual errors in UV-MFRSR cosine or shadowing corrections will manifest themselves as systematic ln(V0) changes with solar zenith angle. Thus, a longer calibration period permits a better estimate of possible systematic errors.

Unexplained V0 results Diurnal V0 changes Examining diurnal trends in V0 data provides insight into possible systematic calibration errors and yields a tool for checking consistency between AERONET-CIMEL ext and UV-MFRSR voltage measurements. Indeed, for perfect measurements, V0 should remain constant during the day regardless of any changes in atmospheric conditions, solar elevation, and azimuth. For example, any systematic residual errors in UV-MFRSR cosine or shadowing corrections will manifest themselves as systematic ln(V0) changes with solar zenith angle. Thus, a longer calibration period permits a better estimate of possible systematic errors.

Comparison statistics: (1) Extinction  (all clear sky cases ~10,000)   < 0.02 (daily rms differences) for all clear sky days   < 0.01 (daily rms differences) for  < 0.5 Daily aerosol extinction optical thickness rms differences between UV-MFRSR and AERONET CIMEL measurements at 368nm. Inferred values of the effective UV imaginary refractive index were used for comparisons of aerosol single scattering albedo, 368, at 368nm with AERONET retrievals at 440nm, 440, using the Dubovik and King algorithm25. The measured differences in absorption between 368nm and 440nm might suggest the presence of selectively UV absorbing aerosols5 or interference from gases other than ozone34. In our opinion, the 368 results we reported do not allow confident explanation of the causes of apparent larger absorption at 368nm compared to 440nm. Continuing co-located measurements at GFSC location is important to improve the comparison statistics, but conducting these measurements at different sites with varying background aerosol conditions is also desirable. The differences in k were consistent with lower  values in UV (368=0.89-0.92 compare to 440=0.93-0.95). This suggests that  spectral dependence in the visible (lower  at longer wavelengths)27 flattens out and even reverses in UV. However, it is emphasized that, except for SZA>70o , the  retrieved at 368nm and 440nm are within the range of overlap of retrieval uncertainties It should be mentioned for AERONET wavelengths, that the  increases with decreasing  for fine mode smoke or pollution aerosol27. Therefore, the extrapolated differences in 368 (predicted by AERONET) and 368 retrieved by UV-MFRSR may be slightly greater than direct comparisons of 440 to 368. Therefore, it is important in the future to conduct similar measurement in locations affected by desert dust or more polluted sites, where UV aerosol absorption is expected to be higher

but the scatter increases on days with high turbidity (lnV0 ~0 but the scatter increases on days with high turbidity (lnV0 ~0.03 for ext~0.6,,). Calculating daily means of individual V0 values reduces the effect of random errors for transferring the AERONET calibration by the square root of the number of available measurements. Typically there are ~60 daily V0 estimates in winter compared to ~200 in summer, therefore, random error in estimating daily mean <V0> value (shown as a horizontal line in figure ) is reduced by a factor 8 to 20. The effect of random errors was further reduced by removing outlier measurements (with ln(V0) outside of +/-3lnV0 of the <ln(V0)>) and iteratively re-calculating <V0>. We found that removing less than 5% of outliers reduces lnV0 by half on both clear and turbid days (in figure lnV0 are shown for 3 iterations). AOT comparisons