SEARCH

SEARCH BY CITATION

Keywords:

  • Ultraviolet;
  • satellite;
  • UV

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Information on ultraviolet (UV) radiative fluxes is needed for public safety, understanding biodiversity, and for chemical transport modeling. Space-based observations can provide homogeneous and systematic estimates of the UV flux over large regions. In the past, UV flux estimates have been made from polar orbiting satellites; such estimates lack information on diurnal variability that can result in significant errors in UV dose (diurnally integrated UV flux). An algorithm has been developed to estimate diurnally varying spectral UV flux at the surface based on information from geostationary satellites (cloud amount, surface albedo and aerosols) and from polar orbiting satellites (ozone). Algorithm evaluation is done by comparison with ground-based observations made between January 1998 and December 2000 over eighteen stations of the United States Department of Agriculture (USDA)'s UV monitoring network. A good agreement between ground-based observations and satellite estimates is found with a mean bias (satellite − ground) of +3.5% for all-sky (cloudy + clear) cases. A negative mean bias of the same magnitude is found for clear-sky cases. Root mean square (RMS) differences are 25% and 14% for all-sky and clear-sky cases, respectively. Using simulations, it is shown that when only one observation near noontime is used to estimate UV dose, errors in the range of −61% to 48% can result, depending on cloud conditions. The RMS difference is 9% and it increases to 13% when off-noon hour (±2 hrs) observations are used to estimate the UV flux over Queenstown, MD.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Ultraviolet (UV) radiation plays an important role in the earth's biosphere by having both harmful and beneficial effects. Harmful effects include increase in cataract incidence, melanoma, sunburn, immune suppression, and photo-aging [Walterscheid et al., 2006; Lucas et al., 2006]. For a 1% decrease in ozone, it is estimated that 100,000 to 150,000 additional cases of cataract would develop [UNEP, 1998]. A World Health Organization (WHO) report finds that excessive UV radiation causes a loss of 1.5 million disability adjusted life years and 60,000 premature deaths globally per year [Lucas et al., 2006]. In the same report, it is shown that 90% of the global burden of disease from melanoma and other skin cancers are due to exposure to UV radiation. Experimental studies have demonstrated that UV rays can reduce crop yield and quality [Teramura et al., 1990; Kakani et al., 2003; Biggs et al., 1981]. Beneficial effects include synthesis of Vitamin D in the human body and treatment of psoriasis [Grant, 2006]. Many birds and insects can see UV rays and their reflectance plays an important role in their social life [Dresp et al., 2005; Mougeot and Arroyo, 2006]. UV radiation also plays an important role in atmospheric chemistry; for instance, production of dimethyl sulfide (DMS), hydroxyl (OH) radicals, tropospheric ozone and ozone precursors are highly affected by availability of UV radiation [Martin et al., 2003; He and Carmichael, 1999; Liao et al., 1999; Dickerson et al., 1997; Hefu and Kirst, 1997].

[3] Effects of UV radiation are wavelength dependent. The relative effectiveness (action spectra) of UV radiation to cause DNA damage, melanoma, and plant growth inhibition as a function of wavelength is shown in Figure 1. For a typical atmospheric condition, surface reaching UVB flux (wavelength range 280 to 320 nm) is one order of magnitude lower than UVA (320 to 400 nm) flux but its effectiveness to cause DNA damage and Erythema is two to three orders of magnitude higher [de Gruijl et al., 1993; McKinlay and Diffey, 1987]. Overall effectiveness of surface reaching UV flux can be accounted for by multiplying the spectral UV flux with a suitable weighting function and integrating it over wavelength.

image

Figure 1. Relative responses for plant growth inhibition [Flint and Caldwell, 2003], skin cancer [de Gruijl et al., 1993], standard erythemal spectra [McKinlay and Diffey, 1987], and ozone absorption spectra [Burrows et al., 1999].

Download figure to PowerPoint

[4] Surface reaching UV flux depends on a variety of factors such as solar zenith angle, column ozone, cloud cover, aerosol optical depth, cloud optical depth, and surface reflectance. These factors are highly variable in space and time and add to the complexity of regional and temporal variation in UV flux. Ground-based observations represent reliable measures of the UV flux. However, being point measurements their representativeness of larger scales is limited. Only satellite-based observations can provide homogeneous and systematic estimates over large regions. Notable among satellite-based estimates of Erythemal UV Irradiance are those from the TOMS and GOME instruments flown on polar orbiting satellites [Eck et al., 1995; Krotkov et al., 1998, 2001; Lubin et al., 1998; Herman et al., 1999; Peeters et al., 2000]. One of the limitations of these satellites is their inability to provide the diurnal variation of the UV flux. Lubin et al. [1998] estimate diurnal variability using monthly mean cloud optical depth and found that the greatest variability in surface UV flux within a given climate zone is induced by clouds. Cloud cover, being highly variable at diurnal scales, can introduce errors in estimates of diurnally integrated UV flux known as UV dose. Bugliaro et al. [2006] have studied the effect of sampling interval on estimates of UVB daily dose using simulations and synthetic cloud cover. They have found more than 35% decrease in maximum uncertainty in five out of six cases, when one hour sampling interval (a case with geostationary satellites) was simulated by a single overpass per day (the case with polar orbiting satellite). Geostationary satellite observations capture diurnal variation but have limited spectral channels, limited spatial coverage and no onboard calibration of the visible channel.

[5] A more comprehensive approach for obtaining estimates of UV fluxes is to merge observations from multiple satellites. Such an approach has been tried by Verdebout [2000, 2004] and by Wuttke et al. [2003] for European regions using cloud information from Meteosat and ozone retrievals from GOME and TOMS. The National Oceanic and Atmospheric Administration (NOAA) operates a system of Geostationary Operational Environmental Satellites (GOES), which carry an imager and an atmospheric sounder. The imager provides information in five spectral channels at half-hourly time intervals over the continental USA and a large portion of the Atlantic and Pacific oceans. The present work explores the possibility to estimate surface UV flux from GOES-8 information on cloud amount, surface albedo and aerosols coupled with ozone estimates from TOMS carried onboard polar orbiting satellites. Details on algorithm development, input data processing, and error budgets are presented in section 2. Algorithm validation against ground-based observations is discussed in section 3. Initial results of UV dose and achievable improvement by inclusion of information on diurnal variation of atmospheric parameter are discussed in section 4. Summary is presented in section 5.

2. Methodology

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Background

[6] To estimate surface level UV flux, information on the top of the atmosphere (TOA) incident UV flux and on the transmission of the atmosphere is needed. Since the solar flux is relatively constant, it is the changes in atmospheric transmission, which gives rise to spatial and temporal variation of the surface fluxes. Atmospheric transmission can be estimated by solving the radiative transfer equation for known values of atmospheric variables. A real challenge lies in using satellite observations to estimate such atmospheric variables needed as input in the radiative transfer model. Numerical solution to radiative transfer equations is a computationally intensive and time-consuming process. An alternative approach is to establish a direct relationship between an observable quantity and transmission or surface flux (known as parametric equations) using regression analysis, hence bypassing detailed solution to radiative transfer equation. Li et al. [2000] have followed this approach to estimate UVB flux from TOMS observations. Such equations employ empirical constants and thus limit the range of applicability. Another approach is to use lookup tables computed off-line for discrete values of important variables such as solar zenith angle, ozone amount, cloud optical depth, aerosol optical depth [Verdebout, 2000]. These can be used to estimate the transmission for the actual value of each variable by interpolation Such approach is followed in this study.

[7] The primary shortcoming in most methods to estimate daily doses of UVB is the lack of representation of the diurnal variability of clouds. To estimate surface UV fluxes, Verdebout [2000, 2004] and Wuttke et al. [2003] used geostationary satellite images from Meteosat for information on the diurnal variability of clouds, and polar orbiting satellites such as TOMS, TOVS and GOME for ozone information. Application of their algorithm is restricted to European regions. In this study, we combine capabilities of polar orbiting satellites to provide information on ozone and geostationary satellites to provide information on the diurnal variation in clouds for estimating UVB fluxes over the United States. The feasibility to implement such an approach is due to an ongoing activity at the University of Maryland (UMD) and NOAA/NESDIS to estimate shortwave (SW) fluxes in five spectral bands from GOES satellites using a lookup table approach. These are produced at hourly time scales over an area bounded between 25 N and 50 N latitudes and 70 W and 125 W longitudes [Pinker and Laszlo, 1992; Pinker et al., 2002, 2003, 2007; Li et al., 2007]. Byproducts include cloud amount, cloud and aerosol optical depth and surface albedo. Since the primary objective of the UMD and NOAA/NESDIS activity is to estimate shortwave fluxes, the parameterization used in the UV part of the spectrum was not sufficiently detailed for estimating UVB. In the present study developed is a new parameterization for deriving UVB and used are relevant by-products from the SW model such as cloud amount, cloud and aerosol optical depth, to estimate the diurnal variability of the UVB. Major changes in the new UV parameterization include the fine spectral resolution in these wavelengths, new transmission tables, inclusion of surface elevation effects, increased number of aerosol and cloud types, and a new interpolation scheme. TOMS's ozone values which are on a coarse grid are linearly interpolated to a 0.5 deg grid to be used with the present algorithm. Default output of the model is total and diffuse UV flux at 5 nm wavelength intervals between 280 and 400 nm. The output module provides option for integration to get erythemal UVB flux. More details on the new algorithm are provided in the following section.

2.2. Algorithm Development

[8] Lookup tables for atmospheric transmission for discrete values of solar zenith angle, aerosol and cloud optical depths, ozone, water vapor and surface elevation are prepared using the Santa Barbara Discrete ordinate Atmospheric Radiative Transfer model (SBDART) [Ricchiazzi et al., 1998]. SBDART accounts for absorption using low-resolution band models originally developed for the LOWTRAN-7 atmospheric transmission code [Pierluissi and Peng, 1985] and radiative transfer equations numerically integrated using the discrete ordinate method of Stamnes et al. [1988]. It uses Mie scattering code to get requisite input parameters for clouds [Wiscombe, 1980]. Extensive information on SBDART and its validation in the shortwave and longwave spectral regions are presented by Ricchiazzi et al. [1998]. The UV spectral output of this study based on SBDART is compared to results of van Weele et al. [2000] who evaluate the performances of 12 different radiative transfer models in the UV spectrum for six typical atmospheric conditions against ground-based observations. The six cases differ in column ozone amount, aerosol optical depth, solar zenith angle, surface albedo and surface elevation. The same sets of inputs are used by all the models and two sets of mean values are computed: one for models that assume plane-parallel geometry and the other for models that use pseudo-spherical approximation. In four out of the six cases, mean values computed by van Weele et al. [2000] are within 13% from observations over the entire UV spectral region. Using the same set of inputs as were used in above study, SBDART was run resulting in transmissions that are within 0.5% from the mean values derived by van Weele et al. [2000] and do not exceed more than 2% for any wavelength (for the plane parallel case). A comparison of the transmission at 310 nm is shown in Figure 2.

image

Figure 2. Comparison of transmission for six different atmospheric conditions between SBDART output and mean values of van Weele et al. [2000]. See van Weele et al. [2000] for more details on input.

Download figure to PowerPoint

[9] The new lookup table contains data for atmospheric transmission and reflection over a dark surface (zero reflectivity) for a range of discrete values of solar zenith angles, aerosol optical depths, cloud optical depths, total column ozone, and surface level pressures. Major features of the lookup table are described in Table 1. SBDART allows the user to specify an extraterrestrial solar UV spectrum. Currently the LOWTRAN-7 solar spectrum is used. The spectral resolution in the lookup tables is kept at 5 nm intervals. The selection of the spectral resolution is based on the structure of the absorption and action spectra and computational restrictions such as processing time and memory requirements. Since no fine structure in ozone spectra in the UV is present, specific values can be estimated from a coarser resolution. Similar to the absorption spectrum of ozone, action spectra are also smooth but a few of them have a sharp drop in value between wavelengths 300 to 328 nm (Figure 1). The UV flux below 280 nm is zero for almost all atmospheric conditions and hence, the spectral range was kept between 280 and 400 nm.

Table 1. Lookup Table Characteristics
 No. of ValuesRange
Cosine of solar zenith angle90.2 to 1
Ozone (cm)30.05 to 0.5
Aerosol optical depth (0.55 μm)60 to 1.5 (boundary layer)
Aerosol type10see Hess et al. [1998]
Aerosol vertical profile10–2 km (as described above)
2–12 km (as described above)
12–20 km background stratospheric
20–70 km meteor dust
Cloud optical depth (0.55 μm)120 to 200
Cloud type2water and ice
Surface pressure (mbar)41013 to 650
Water vapor (cm)40.5 to 5
Wavelengths (μm)290.28 to 0.4 step 0.005 (spectral values)
0.4 to 0.7 step 0.1
0.7 to 4

[10] Ten aerosol types as given by Hess et al. [1998] are used in preparing the lookup tables. Aerosols differ in single scattering albedo, asymmetry parameter and spectral dependence of extinction. Exponentially decreasing aerosol extinction profile is used with different scale height in the boundary layer, free troposphere and stratosphere. Background stratospheric aerosol type is used for the altitude range between 12 and 20 km with fixed layer optical depth at the 0.55 μm wavelength equal to 1.744e-03 and meteor dust type for the layer between 20 and 70 km with fixed layer optical depth equal to 3e-3. The aerosol optical depth in the boundary layer is variable and a lookup table is prepared for six aerosol optical depths between 0 and 1.5 at the 0.55 μm wavelength. Currently only two aerosol types are considered in retrieving fluxes, one for marine and one for continental environment. However, the algorithm is generic in nature and with future improved aerosol information distinction can be made between additional aerosol types.

[11] The cloud types considered for simulations are ice and water clouds. A sensitivity study was performed to assess the effect of cloud droplet effective radius, cloud base height, and cloud physical thickness; it was found that the dependence of surface level fluxes on cloud height, physical thickness and effective droplet size is insignificant when cloud optical depth is held constant (Figure 3). Hence fixed cloud height (2 to 5 km for water and 9 to 11 km for ice cloud) and effective droplet radius (10 μm for water and 23 μm for ice cloud) are used. Twelve cloud optical depths are considered between 0 and 200, where eleven values are between 0 and 80.

image

Figure 3. Sensitivity of erythemal UVB transmission to cloud parameters (cloud optical depth, effective droplet size, cloud-base height). Other parameters are held constant (solar zenith angle (30 deg), column ozone amount (300 DU), aerosol optical depth (0.1), surface albedo (0.03), and surface pressure (1013 mbar)).

Download figure to PowerPoint

[12] A sensitivity study was performed for different atmospheric profiles, namely, the US standard 1962, tropical, mid latitude summer, mid latitude winter, sub arctic summer, and sub arctic winter [McClatchey et al., 1972]. It was found that the dependence of surface UV flux on atmospheric profiles is insignificant when column integrated values for ozone are kept constant. Atmospheric profiles affect only relative vertical distribution of ozone and water vapor since profiles are scaled to match specified column integrated value. Hence only one atmospheric profile (US standard 1962) is used for generating the lookup tables over a dark surface. Effect of surface reflectivity is modeled using Pinker and Laszlo [1992, equations (1) to (3)].

2.3. Input Data and Processing

[13] The UV module is designed in a way that it can be integrated with the UMD/Shortwave Radiation Budget (SRB) algorithm for the entire solar spectrum. Cloud optical depth, cloud type, cloud fraction, aerosol optical depth, surface albedo, solar zenith angle, and sun-earth distance, which are inputs to the UV module, are determined in the UMD/SRB algorithm (Version 2.1: [Li et al., 2007; Pinker et al., 2007]: http://www.atmos.umd.edu/~srb/gcip/webgcip.htm). Clear-sky, cloudy-sky and clear-sky composite radiances, cloud fraction, cloud classification, snow cover and monthly mean water vapor are provided at 0.5 × 0.5-degree grids. Cloud optical depth, aerosol optical depth, and surface albedo are also available from the UMD/SRB model, however, these were re-calculated with the new lookup tables prepared with the SBDART code to make them consistent with the SBDART-based UV lookup tables (the original UMD/SRB lookup tables were prepared with the Delta-Eddington code).

2.4. Pre-processing of Input Data for UV Flux Estimates

[14] Top of the atmosphere radiances are obtained by multiplying instrument counts with a calibration constant. The GOES Imager is calibrated vicariously in the absence of an on-board calibration facility and calibration procedure accounts for sensor degradation over time. NOAA/NESDIS uses a radiometrically stable reference site in the Sonoran Desert (34°N, 114°W) for calibration of the GOES imager [Rao et al., 1999]. NASA uses time and space collocated images from research satellites with onboard calibration facility for calibrating GOES visible channel [Minnis et al., 2002]. However, the choice of calibration method does not make a significant difference in estimating cloud cover or shortwave fluxes as shown by Wonsick et al. [2006] because the algorithm relies on the relative differences between clear-sky composite and instantaneous radiances and not on the absolute values of radiances. Clear-sky composite images/radiances for a given month are obtained from multiple days by conservatively selecting cloud free images. They represent radiances for clear-sky conditions with minimum aerosol and atmospheric contribution. Cloud identification is done using coupled cloud and snow detection algorithms [Li et al., 2007].

[15] Surface albedo is estimated using radiative transfer equations, clear-sky composite and climatological aerosol optical depths (AOD) [Liu et al., 2005]. Thus obtained surface albedo is used to estimate aerosol and cloud optical depth from instantaneous top of the atmosphere radiances. The underlying assumption is that surface albedo has less temporal variability on a day to day basis; hence, variations of radiances at the top of the atmosphere are due to variation in atmospheric factors such as cloud, aerosol, ozone or water vapor. As described in detail by Pinker and Laszlo [1992], cloud and aerosol optical depths are computed by matching computed and GOES-derived top of the atmosphere broadband albedos. The latter are obtained from the narrowband GOES radiances after applying appropriate spectral and angular transformations [Zhou et al., 1996; Pinker et al., 2003].

[16] Water vapor values are taken from the National Centers for Environmental Prediction (NCEP) outputs [Kanamitsu et al., 2002]. Ozone values are obtained from the Total Ozone Mapping Spectrometer (TOMS's) data (http://toms.gsfc.nasa.gov/) maintained by the NASA Goddard Space Flight Center (GSFC). TOMS is a satellite instrument successively launched by NASA onboard Nimbus-7, Meteor-3 and Earth Probe (EP). It provides near continuous global coverage of daily ozone amounts from 1978 to present. Quality controlled level-2 TOMS data are given over a 1° × 1.25° lat-lon grid. These data were re-gridded here to 0.5° × 0.5° latitude/longitude grids by linear interpolation to match the cloud products from GOES. The error in TOMS ozone values is between 3% and 5% depending on the solar zenith angle [McPeters et al., 1998]. Surface elevation data are obtained from the U. S. Geological Survey web-site (http://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html). The elevation data known as GTOPO 30 are available at 30 arc sec resolution. They are averaged over 0.5° latitude longitude grids and converted into pressure units using a standard atmosphere.

[17] Separate calculations for clouds and aerosols are carried out for each grid point and all-sky fluxes are obtained by taking a weighted average for cloudy and clear-sky pixels. The output of the interpolation module are total (direct + diffuse) and diffuse fluxes at 5 nm spectral intervals in the UV range between 280 and 400 nm for clear-sky, cloudy-sky and all-sky conditions. The estimated spectral UV flux is weighted with the Commission Internationale de I'Eclairage (CIE) erythema spectra [McKinlay and Diffey, 1987] and integrated between the wavelength range 280 to 330 nm and compared with ground-based observations. It should be noted that many different wavelength ranges are used to describe UVB in the literature depending upon the end-use of UVB radiation and action spectra relevant to that field. CIE defines the wavelength range 280 to 315 as UVB radiation while the wavelength range 280 to 320 nm is also quite common in dermatological studies.

2.5. Theoretical Error Estimates

[18] Accuracy of a model is limited by various sources of error such as in input parameters, radiative transfer scheme, interpolation, and rounding off. The largest of them is the error in input parameters. Theoretical error estimates are made by varying respective atmospheric parameters within their error limit and calculating the difference in model output. A typical error estimate for TOMS's ozone value is 3%, which can lead to about 3% error in erythema UVB (E-UVB) flux. The error in aerosol optical depth has (AOD) less effect on the total E-UVB flux except for highly absorbing aerosols; about 10% error in AOD leads to 1% error in E-UVB flux for most aerosol types. The relationship between errors in cloud optical depth (COD) and corresponding errors in UVB flux is complex. For a typical COD of 5, 10% error in COD leads to about 2% error in E-UVB flux, whereas for COD of 25, 10% error in COD leads to about 5 to 6% error in E-UVB flux. Surface albedo in UV spectrum is very low (∼0.03) hence it has less effect on total UVB flux (unless there is snow on the ground). In case of snow, a 10% error in albedo can lead to about 6% error in E-UVB flux. Error due to interpolation and rounding off is less than 0.2%.

3. Validation of UVB Fluxes

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

[19] Model estimates of UVB are evaluated against ground-based observations obtained from the US Department of Agriculture (USDA) UVB monitoring stations that started operation in 1994 [Bigelow et al., 1998]. Currently, 36 stations are operational and distributed over the United States including Hawaii and Alaska. Some are of research quality and some of climatological value. The network includes a station in Canada and New Zealand. Erythemal UVB fluxes are measured using UV-1 pyranometer manufactured by Yankee Environmental Systems. Measurements are made every twenty seconds and averaged over three minutes by the on-board processor. The network UV-1 pyranometers are calibrated against a triad of standard UVB instruments that are maintained by the Surface Radiation Research Branch (SRRB)'s central UV calibration facility (CUCF). The standard instruments are periodically calibrated outdoors by comparing their broadband measurements to the integrated output of UV spectroradiometers [Lantz et al., 1999]. UV-1 pyranometer data are given as erythemal UV irradiance, the total measured UVB flux convoluted with the erythemal action spectrum. E-UVB irradiance is calculated by assuming column ozone value equal to 300 Dobson units (DU). Overall radiometric accuracy is ±2.5% for solar zenith angles less than 80 degrees [Bigelow et al., 1998]. However, deviation of the actual ozone value from 300 DU can introduce additional errors, which can be as high as 15% depending upon ozone amount, solar zenith angle and season [Lantz et al., 1999].

[20] Data from eighteen stations are used for evaluation (Figure 4 and Table 2). Spectral output of the model after multiplying with CIE erythemal weighting function [McKinlay and Diffey, 1987] is integrated between wavelength 280 and 330 nm for comparison with ground-based observations. The ground-based observations are averaged over 30 minutes centered on time of the instantaneous model estimates. Figure 5a shows a comparison for all-sky conditions for the eighteen stations during January 1998 to December 2000. Color represents the population of scatter points. The mean bias between satellite estimates and ground observations is 2.6 mW/m2 and the RMS difference is 18.3 mW/m2. The correlation coefficient for linear least square fit for 98 thousand data points is 0.96. The slope of the linear least square fit is 0.98 and the intercept is 4 mW/m2. A similar comparison for clear-sky conditions is shown in Figure 5b. Clear sky is selected based on cases when the satellite-based cloud fraction in a particular grid is zero. The scatter in data points is significantly reduced for the clear-sky case; however, there is an underestimation in the satellite-based values. For clear sky the mean bias is −3.5 mW/m2 and the RMS difference is 13.2 mW/m2. The comparison of satellite estimates against ground observations has inherent limitations. Satellites see much larger area as compared to ground instruments, which can result in difference in cloud fractions, aerosol optical depth, ozone amount, surface albedo, and elevation in field of view giving rise to difference in fluxes between the two methods. Since clouds have higher spatial and temporal variability compared to other parameters affecting E-UVB flux, they tend to produce more variability compared to clear-sky cases. While the RMS difference can be explained in terms of difference in viewing geometry of ground and satellite observations, mean bias differences require additional explanation. For the eighteen stations the mean bias ranges from −3.9 mW/m2 to +8 mW/m2 for all-sky conditions and −13 mW/m2 to +7 mW/m2 for clear-sky conditions (Table 2). The mean bias is of the order of 3.5 %, when all the stations are merged for both clear and cloudy-sky conditions. For clear-sky cases, there is an overall tendency of underestimation since only three stations have positive mean biases. Possible causes for this: Error in extraterrestrial flux, difference in filter response function of ground instruments and CIE erythemal weighting function used for satellite estimates; deviation of ozone values from 300 DU; error in modeling of spectral properties of aerosols, surface albedo, and ozone absorption.

image

Figure 4. USDA stations used for the validation of satellite estimates of erythemal UVB.

Download figure to PowerPoint

image

Figure 5. Comparison of satellite estimated erythemal UV flux with ground observations from the USDA UV Network for (a) all-sky condition and (b) clear-sky condition.

Download figure to PowerPoint

Table 2. Comparison of Satellite Estimates With USDA Stations for UV Flux Monitoring
Ground StationTime PeriodLatitudeLongitudeCorrelation Coefficient All SkyMean BiasRMS
All SkyClear SkyAll SkyClear Sky
Castolon Site at Big Bend National Park, Panther Junction, TexasJanuary 2000 to December 200029.13 N103.51 W0.97−0.4−1219.713.6
LSU Central Research Station, Baton Rouge, LouisianaJanuary 1999 to December 200030.35 N91.16 W0.947.6−325.813.5
Jornada Experimental Range, Las Cruces, New MexicoApril 1999 to December 200032.61 N106.74 W0.96−3.9−1323.14.8
Desert Research and Extension Center, Holtville, CaliforniaJanuary 1999 to December 200032.80 N115.44 W0.982.3714.720.2
Bledsoe Research Farm, Griffin, GeorgiaMay 1999 to December 200033.18 N84.41 W0.958.−0.121.916.9
UC Davis Climate Station, Davis, CaliforniaMarch 1999 to December 200038.52 N121.76 W0.981.20.313.313.3
Wye Research and Education Center, Queenstown, MarylandApril 1998 to December 200038.91 N76.14 W0.963.−5.417.8
Ecology Research Center, Oxford, OhioJanuary 1998 to December 200039.52 N84.71 W0.952.2−7.218.910.7
Bondville Road Station, Bondville, IllinoisJanuary 1999 to December 200040.04 N88.36 W0.964.2−6.517.813.2
Central Plains Experimental Range, Nunn, ColoradoNovember 1999 to December 200040.79 N104.75 W0.951.3−4.322.312.6
Agricultural Research and Development Center, Mead, NebraskaApril 1998 to December 200041.13 N96.48 W0.965.−4.218.49.6
Utah Climate Center, Logan, UtahSeptember 1999 to December 200041.66 N111.89 W0.965.35.420.214.7
New York State Agricultural Experiment Station at Geneva, Geneva, New YorkJanuary 1999 to December 200042.87 N77.02 W0.951.6−8.516.211.2
Proctor Maple Research Center, Burlington, VermontSeptember 1998 to December 200044.53 N72.85 W0.954.7−6.816.8.9
Lake Dubay, Dancy, WisconsinSeptember 1999 to December 200044.70 N89.76 W0.952.−6.714.28.6
University of Michigan Biological Station at Douglas Lake, Pellston, MichiganMay 1998 to December 200045.55 N84.66 W0.950.6−6.716.58.9
Albion Field Station, Pullman, WashingtonJune 1998 to December 200046.75 N117.18 W0.97−3.2−415.18.8
North Central Research and Outreach Center, Grand Rapids, MinnesotaJanuary 1999 to December 200047.18 N93.53 W0.952.7−416.47.5
All Stations1998–2000------0.962.6−3.518.313.2

4. Daily UV Dose

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

[21] Daily UV dose is a time integrated E-UVB flux over a period of one day. Estimates of UV dose from polar orbiting satellites are made by assuming diurnally constant atmospheric variables. This assumption can lead to errors, particularly for days with significant variability in cloud cover. E-UVB fluxes for one such a day (April 12, 1999) over Queenstown, Maryland are shown in Figure 6, where the solid line shows the E-UVB flux from ground observation and circles shows flux estimates using GOES data. In order to evaluate the improvement in UV dose estimates from the GOES data, simulations of daily UVB flux were performed using hourly observations from GOES as opposed to a single observation. Percentage difference of E-UVB dose estimated from a single observation around 11:15 am and hourly observations is shown in Figure 7a; it varies from −61% to 48% for single day. A positive difference means the single observation has higher E-UVB dose than that obtained from hourly observations. Moving cloud systems are well accounted for by frequent observations, while a single observation can miss them at one place but over account for them at another. The difference reduces to −25% to 25% when averaged over one month but affected area increases significantly (Figure 7b). A similar simulation was performed for one location (Queenstown, MD) for an extended period (January 1998–December 2000). First, utilized are hourly observations from GOES (henceforth, case1). Subsequently, used is only one observation around local noon (17:15 UTC), one two hours before noon (15:15 UTC) and one two hours after noon (19:15 UTC) (henceforth case2, case3 and case4). These times approximately correspond to the local overpass times for Earth Probe TOMS (EPTOMS), and MODIS-Terra and MODIS-Aqua (16:16, 15:30 and 18:30 UTC). The difference between case1 and case2 is as high as 40% with an RMS difference equal to 9%. However, the mean bias between case1 and case2 is only 1.6%, which reconfirms the conclusion of Herman et al. [1999] that the effect of diurnal cloud variability is reduced when UV dose estimates are averaged over a longer time scale. When using only one observation to estimate UV dose, noontime observations are most important. Case3 and case4 are two hours off-noon; it was found that RMS difference between case3 and case1 (or case4 and case1) is 4% higher than the difference between case2 and case1. Similar results were found by Bugliaro et al. [2006] who used synthetic cloud diurnal variability to study their effect on UV flux. Since major contribution in UV dose comes from the UV flux around noontime, errors in atmospheric variables such as cloud optical depth will have larger impact when made during noontime than off-noontime.

image

Figure 6. Erythemal UVB flux observed at Queenstown, MD on 12 April 1999. Solid line: ground-based observations; circles: estimates from GOES data; dashed and dotted lines: estimates using one GOES observation at 10:15 hrs only and at 14:15 hr only, respectively, and assuming them constant over the day.

Download figure to PowerPoint

image

Figure 7. (a) Percentage difference in UV erythemal dose estimates using single observation around 11:15 and hourly observations for 12 April 1999. (b) Same as Figure 7a but monthly mean values for April 1999. Positive difference means single observation estimates are high. (c)–(f) Daily UV erythemal dose (kJ/m2), cloud optical depth, column ozone amount (DU), and cloud fraction over United States on July 1, 1998.

Download figure to PowerPoint

[22] In Figure 7c, a sample of UV dose on 1st July 1998 over the USA is shown. Missing ozone values were interpolated by triangulation to estimate UV dose on that day. Color range from violet to red shows UV dose ranging from 0 to 6.5 kJ/m2. Cloud optical depth, column ozone amount and cloud are shown in Figures 7d7f. Effect of clouds can be seen over Northern Mexico and Central Southern USA. The overall UV dose pattern is the result of combined effect of clouds, ozone, sun-earth geometry and surface elevation. Low UV dose in north-east USA and the east coast of Canada is due to high column ozone values in this region. However, the lowest UV flux value does not coincide with the highest ozone value in the north-east; instead, the lowest value is north of the ozone minimum reflecting the combined effects of ozone and clouds. North to south gradient is due to sun-earth geometry. Western USA has higher UV dose compared to eastern USA for the same latitude due to higher surface elevation and lower cloud amount.

5. Summary

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

[23] An algorithm to estimate diurnal variation of surface reaching UV flux has been developed for use with GOES information on cloud amount, surface albedo, aerosol, and ozone estimates from polar orbiting satellites. The algorithm provides spectral estimates of the UV flux at 5 nm intervals and differentiates between total and diffuse flux, making it suitable for a wide variety of scientific applications.

[24] The UV flux estimates are evaluated against ground-based observations made at eighteen stations of USDA's UVB monitoring network distributed over the USA. The mean bias between ground and satellite estimates of UVB flux is 3.5%. The RMS difference is 25% for all-sky conditions and 14% for clear-sky conditions. A large reduction in RMS difference for clear sky shows the importance of cloud spatial and temporal variability when comparing satellite estimates with ground observations.

[25] A simulation study for one day over the United States shows that the UV estimates from a single observation around noontime tend to produce errors in daily UV dose between −61% and 48%. A similar simulation is carried out for one location (Queenstown, Maryland) for an extended period (January 1998 to December 2000); an RMS difference of about 9% is observed. The RMS difference increases to 13% when observations only from morning or evening (local noon ±2 hrs) hours are used.

[26] We see significant effects of clouds and surface elevation on surface downwelling UVB flux in addition to those from the sun-earth geometry and ozone amount. The south and west USA are likely to receive more UV radiation in summer because of the lower solar zenith angle, higher surface elevation in the west in some places and reduced cloud cover.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

[27] The authors wish to thank all those involved with the USDA UV monitoring network, in particular, James Slusser for generously providing data for validation. This research was conducted within the Cooperative Institute for Climate Studies, NOAA Cooperative Agreement NA17EC1483 under grant B2NS1GG from the National Oceanic and Atmospheric Administration (NOAA), Office of Systems Development to the University of Maryland. NCEP Reanalysis data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov/.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Validation of UVB Fluxes
  6. 4. Daily UV Dose
  7. 5. Summary
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrd14355-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
jgrd14355-sup-0002-t02.txtplain text document2KTab-delimited Table 2.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.