Ultraviolet Index Forecasts Issued by the National Weather Service # C

Ultraviolet Index Forecasts Issued by the National Weather Service # c

Craig S. Long,* Alvin J. Miller,* Hai-Tien Lee,+ Jeannette D. Wiid,+ Richard C. Przywarty,* and Drusilla Hufford@

ABSTRACT

The National Weather Service (NWS), in collaboration with the Environmental Protection Agency (EPA), now issues an Ultraviolet (UV) index forecast. The UV index (UVI) is a mechanism by which the American public is forewarned of the next day's noontime intensity of UV radiation at locations within the United States. The EPA's role in this effort is to alert the public of the dangerous health effects of overexposure to, and the accumulative effects of, UV radiation. The EPA also provides ground-level monitoring data for use in ongoing verification of the UVI. The NWS estimates the UVI using existing atmospheric measurements, forecasts, and an advanced radiative transfer model. This paper discusses the justification for a forecasted index, the nature of UV radiation, the methodology of producing the UVI, and results from verifying the UVI. Since the UVI is an evolving product, a short discussion of necessary improvements and/or refinements is included at the end of this article.

1. Introduction war, travel to vacation spots became quicker and more affordable, and fashion permitted wearing less and The sun habits of people prior to World War II less, thus exposing more of the human skin area to were much different than the sun habits of people ultraviolet (UV) radiation. At the same time, socio- today. Leisure time was at a minimum. Transporta- economic influences made having a tan desirable. The tion to sunny places was either nonexistent, too time misconception of a tan being "healthy" began during consuming, or unaffordable. Fashion dictated that this time. Consequently, many have spent hours ly- men and women wear hats outside. Around the turn ing in the sun to get a "good" tan. of the century, women often used parasols to block As a result of this increase in exposure (and over- the sun. Pale skin was a status symbol, and a tan des- exposure) to UV radiation over a period of years to ignated one as a member of the working class. When decades, the number of diagnoses of melanoma and vacationing at the beach or just relaxing outside, fash- nonmelanoma skin cancers, as well as cataracts, has ion dictated that people clothe a major portion of the risen dramatically since 1980. This increase has also body. However, as leisure time increased after the been noted in many other countries for similar rea- sons. The American Cancer Society (ACS 1995) estimated that over 800 000 cases of highly curable basal cell or squamous cell cancers would be diag- *NOAA/NWS/NCEP/Climate Prediction Center, Washington, nosed in 1995 in the United States alone. For 1995, D.C. +Research and Data Systems Corporation, Greenbelt, Maryland. the ACS estimates that melanoma, the most serious #NOAA/NWS/Office of Meteorology, Washington, D.C. skin cancer, will be diagnosed in about 34 100 people. @EPA/OAR/Office of Atmospheric Programs, Washington, D.C. Since 1973, the incidence rate of melanoma has in- Corresponding author address: Craig S. Long, Stratospheric creased by about 4% per year (National Cancer In- Analysis Group, National Centers for Environmental Prediction, stitute 1994). Skin cancer is not just "skin deep"—it W/NP53, World Weather Building, Room 808, Washington, DC can be lethal. The ACS further estimated 9300 deaths 20233. E-mail: [email protected] resulting from skin cancers in 1995: 7200 from ma- In final form 23 October 1995. lignant melanoma and 2100 from other skin cancers.

Bulletin of the American Meteorological Society 7 729

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC This number of 9300 deaths translates to greater than awareness campaigns, while the NWS would be re- one death per hour in the United States. Cataracts sponsible for creating and issuing a practical and re- have been determined to cause 53% of blindness cases sponsible UV index. A reduction in the number of worldwide. In addition, studies by DeFabo and treatments for skin cancer and cataract surgeries could Noonan (1983) show that overexposure to the sun can lead to billions of dollars in health care savings. suppress the immune system. In response to these Prevent Blindness America, a group advocating pre- health-related concerns, several countries have initi- ventative measures, cites from existing data up to ated public outreach campaigns to inform the public 1993 that about 1.35 million cataract surgeries are of the dangers of overexposure to the sun. These cam- performed annually in the United States. This costs paigns explain that simple remedial steps can be taken the taxpayers $3.4 billion in Medicare expenses. It is to greatly reduce the risk of overexposure, and they unknown what percentage of these cataract cases are are reinforced by the daily issuance of an index that directly attributable to overexposure to UV radiation, informs the public of the potential intensity of the but clearly there is great potential for reducing costs sun's UV radiation. by very simple measures, such as wearing broad Queensland, Australia, started the first education rimmed hats and protective sunglasses. campaigns on the prevention of skin cancer and the The focus of this paper is to examine the science hazards of overexposure to UV radiation. In the mid- behind the UV index (UVI). To accomplish this, a 1980s, the Australian Radiation Laboratory began brief review of the nature of UV radiation is pre- monitoring UV radiation and broadcasting the day's sented, followed by the methodology by which the UV dosage in "minimum erythemal dosage" units NWS makes its "clear sky" UVI forecast. Then fol- (MED) for all the states' capital cities during the lows a discussion of the seasonal variability of the evening news. Also, in 1987, New Zealand initiated clear-sky UVI for both hemispheres, along with a public awareness campaigns along with issuing "burn methodology for forecasting the effects of clouds times" reports broadcast hourly on the radio. In 1992, on the clear-sky UVI in the United States. Verifica- the Atmospheric Environment Service (AES) of tion statistics of the UVI by surface observations for Canada began issuing their own UV index (Wilson 1993 and 1994 are presented. Finally, alternative 1993), a next-day forecast of UV exposure on a scale methods and improvements to the methodology are of 0-10 (where 10 is the highest value likely in south- discussed. ern Canada). All three countries have had very good success in getting the message to the public about the dangers of being in the sun too long and possible con- 2. Nature of ultraviolet radiation sequences to the skin, eyes, and immune system over a prolonged period of time. Surveys in Australia and Ultraviolet radiation can be divided into three parts Canada (Hill et al. 1993, personal communication; of the sun's radiation spectrum. Ultraviolet C is char- Decima Research 1993) have shown that people know acterized by wavelengths of less than 280 nm. Al- of the existence of their country's index and have though highly dangerous to plants and animals, this made changes in their sun habits by either avoiding part of the UV spectrum is completely absorbed by the sun during the peak UV radiation hours of the day stratospheric ozone and does not reach the earth's or by slipping on a long sleeve shirt, slapping on a surface. Ultraviolet B ranges in wavelength from 280 hat, and slopping on sun screen lotion, as taken from to 320 nm. Ozone absorbs much of the shorter wave- the Australian SunSmart "Slip, Slap, Slop" campaign. length radiation, but this absorption weakens as In the autumn of 1992, the U.S. Environmental 320 nm is approached. Plants and animals are par- Protection Agency (EPA) approached the National ticularly affected by this part of the UV spectrum. Weather Service (NWS) to develop and generate an Ultraviolet-B effects to humans are reddening of index similar to that of Canada for the United States. the skin (erythema) and reduction of vitamin-D syn- By having an index to act as a beacon, the EPA and thesis in the short term and development of skin can- the Center for Disease Control and Prevention could cer, cataracts, and suppression of the immune system launch a massive public awareness campaign to alert in the long term. The wavelengths of UV-A radiation the American public about the dangers of UV radia- range from 320 to 400 nm. Ozone absorbs very little tion. The EPA would be responsible for interagency of this part of the UV spectrum. Ultraviolet-A radia- coordination and all facets of the public education and tion is needed by humans for the synthesis of vita-

730 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC FIG. 1. The UV spectral irradiances (thin line) from 290 to FIG. 2. The product of spectral irradiances and the erythemal 400 nm at summer solstice, 40°N, and solar noon, with 300 DU action spectrum is plotted. Note that the peak occurs between 308 of ozone overhead. For the same wavelengths but using the and 310 nm. opposite j axis, the erythemal action spectrum (bold line) is plotted.

Figure 2 shows the net product of applying the min D; however, too much UV-A causes photoaging erythemal action spectrum to the irradiance spectrum (toughening of the skin), suppression of the immune shown in Fig. 1. Note that the peak value is near system, and to a lesser degree, reddening of the skin 308 nm. This peak rises (lowers) with decreasing (in- and cataract formation. creasing) amounts of total column ozone. When in- Figure 1 shows a plot of solar spectral irradiances tegrated over the range of 290-400 nm, the resultant (mW nr2 nm-1) at the surface from 290 to 400 nm for erythemal irradiance (mW m~2) or "dose rate" indi- day of the year 172 (22 June) at solar noon, with a cates the instantaneous amount of skin-damaging UV total ozone amount of 300 Dobson units (DU, radiation. This dose rate has been observed to change milli atm cm). Note that the spectral irradiance in- at the rate of about 1.25 ± 0.20% per 1% change in creases by five orders of magnitude between 290 and total column ozone (McKenzie et al. 1991), a concept 400 nm. Although the spectral irradiances drop off known as the Radiative Amplification Factor (RAF). precipitously from 320 to 290 nm, it is in this range The RAF of 1.25 quoted above was determined at that the greatest changes occur in response to changes Lauder, New Zealand (45°S). McKenzie et al. (1991) in total column ozone or in the pathlength that the pointed out that this RAF value tends toward higher sun's light must travel through the atmosphere to values at larger solar zenith angles. However, the reach the surface. These changes are significant. It is amount of UV radiation reaching the surface at these at these wavelengths that plants and animals are most larger solar zenith angles is much less than at smaller sensitive to UV radiation exposure. Also illustrated solar zenith angles. in Fig. 1 is the standard erythemal (or sunburning) ac- A "dosage" value may be obtained by integrating tion spectrum (McKinlay and Diffey 1987) adopted the dose rate over a period of time (i.e., minute, hour, by the Commission Internationale de l'Eclairage day, year). Showing how the UV dose rate varies (CIE) to represent the average skin response over the throughout the day, Fig. 3 represents a typical mid- UV-B and UV-A regions of the spectrum. Note how summer, diurnal curve of erythemally weighted UV the shorter wavelengths are highly sensitive in com- dose rates at 20°N, 40°N, and 60°N, with the same parison to the longer wavelengths. The erythemal amount of ozone overhead. Note the rapid rise (de- action spectrum is a composite of several investiga- crease) during the midmorning (midafternoon) hours. tors' measurements of the response of many differ- However, during the solar noon hour (half hour on ent human skin types to UV radiation and is not rep- either side of solar noon), the curve is relatively flat. resentative of any one skin type. The term "skin type" This curve becomes broader and flatter with lower is meant to depict not only the natural pigmentation noontime values as one approaches the poles and nar- of the skin, but also the likelihood of one's skin to rower and more peaked with higher noontime values either burn or tan. as one approaches the Tropics.

Bulletin of the American Meteorological Society 7 731

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC (1991). The other drivers of the intensity of UV radiation for a particular day are the location's latitude, the time of day, and the amount and thickness of clouds overhead. Other factors that either enhance or attenuate the UV intensity are elevation, surface albedo, tropospheric pollution, and haze. Models exist to incorporate the effects of pollutants and haze on UV radiation reaching the surface (Liu et al. 1991; Michelangeli et al. 1991). However, field observations are lacking to verify these calculations. Ultraviolet radiation does increase with increasing altitude as the depth of the atmosphere and the amount of scattering decreases. It also can be reflected significantly by water, sand, concrete, and snow FIG. 3. Diurnal curves of erythemal dose rates (mW m 2) at 20°N, 40°N, and 60°N at summer solstice, with 300 DU of ozone (Blumthaler and Ambach 1988; Coulson and Reynolds overhead. Note that as higher latitudes are approached, the diurnal 1971; Kondratyev 1969). Each of these may need to curve during the solar noon hour flattens. be accounted for when examining UV observations. The NWS derives ozone amounts from data re- trieved by polar-orbiting satellites and determines 3. NWS methodology cloud amounts from its numerical models. A site's elevation is looked up from a gridded elevation field, To determine the UV radiation at the surface, one given its latitude and longitude. Currently, the UVI can either use instruments at the surface to measure holds the surface UV albedo constant at 5%, which the UV radiation or deduce it from the total column is in agreement with measured UV albedos for the ozone via a radiative transfer model. Due to the lack most common surface types covering the United of adequate, well-calibrated, highly accurate instru- States. (Madronich 1993). The optical depth, which ments within the United States, the NWS uses the lat- is a unitless measure of the opacity of the atmospheric ter approach. The radiative transfer model we make column, is given a constant value of 0.2. A totally use of needs just a few inputs to compute a clear-sky clean atmosphere will have an optical depth of 0.0. (no clouds) determination of the irradiances through- As the amount of absorbing and scattering aerosols out the UV spectrum. Inputs to determine the UV and gases increases, so will the optical depth, possi- spectral irradiances at a specific location include: the bly to as high as 2.5. Future plans to use more accu- total column ozone above that location, the location's rate representations of the UV albedo and the optical latitude, the day of year, and the time of the solar day. depth are addressed later in this paper. The effects and This model uses the latter three inputs to determine the forecast of tropospheric pollution and haze on UV the sun-earth distance and, from this, the amount of radiation still require more research and are not in- solar radiation reaching the top of the atmosphere and cluded in the current UVI. the solar zenith angle. The solar zenith angle determines both the angle of incidence of the UV radiation at the earth's surface and the optical path the UV 4. Sources of ozone data radiation travels through the atmosphere. As the path length becomes greater, so does the absorption and Ozone information needed to determine the UV scattering of UV radiation by ozone and aerosols. radiation at the surface is available to the NWS from Also, as the solar zenith angle increases, the amount several sources. Two instruments on board a number of of UV energy per horizontal area decreases. different satellites currently detect total column ozone. Daily total ozone variations are quite small (±1%) The Solar Backscattering Ultraviolet Ozone Sensor/2 in the Tropics all year round and in the midlatitudes (SBUV/2) instrument obtains vertical profiles of ozone during the summer months. This means that daily within a nadir-viewed footprint. This instrument re- variations in UV radiation amounts under clear-sky sides on board the currently operating NOAA-14 sat- conditions would also vary minimally (~±1.25%), ellite and was also on board the previous NOAA-11 using the RAF determinations of McKenzie et al. and NOAA-9 satellites. Full global coverage of the

732 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC sunlit portion of the earth is achieved as the satellite and temperature fields from "two days ago" to yes- orbits the earth 14 times per day. The data from this in- terday will be applicable for determining the changes strument are processed in a routine mode at the National of the ozone field from yesterday to tomorrow. Two Oceanic and Atmospheric Administration (NO A A). sets of differences are needed to carry out this proce- Because of the small geographical coverage of the dure. The first set is made up of the differences be- nadir-viewed total column ozone data footprints, a tween yesterday and two days ago. These are com- Cressman adjustment scheme is used to incorporate puted for the ozone, temperature, and height fields. the observed data into an existing gridded analysis of Separate regressions for the Northern and Southern the hemispheric ozone data (Nagatani et al. 1977). Hemispheres between the ozone differences and the A second instrument, the TIROS (Television In- height and temperature differences are computed. frared Observation Satellite) Operational Vertical This is done to preserve the seasonal differences be- Sounder (TOVS), provides another source of total tween the two hemispheres. The second set of differ- column ozone data. This instrument is also on board ences is determined from the MRF's forecast fields the current NOAA-14 satellite and was also on board for tomorrow and the analysis fields from yesterday. all previous NOAA satellites. The TOVS instrument Using the regression coefficients and the second set provides horizontally resolved data used to produce of differences, we "forecast" the ozone difference operational products at NOAA. Since the TOVS tem- field between yesterday and tomorrow. Equation 1 perature soundings are already used by the National summarizes the above narrative, where Centers for Environmental Prediction (formerly the

National Meteorological Center) Medium Range • A03 is the forecast change in total column ozone Forecast (MRF) model's analyses, the TOVS ozone from yesterday to tomorrow; data would be a dependent parameter in a compari- • AZ]00 and AZ500 are the changes in geopotential son of the MRF's temperature fields. Since we use a height at 100 and 500 hPa from yesterday to tomor- MRF temperature field in our ozone forecasting row, respectively; and scheme, the SBUV/2 ozone data are the primary in- • AT50 is the change in temperature at 50 hPa from dependent source of ozone data. The TOVS ozone yesterday to tomorrow, and data act as a backup in the event that the SBUV/2 data • a, b, and c are regression coefficients, and d is the become unavailable. This backup procedure was en- regression constant: acted after the NOAA-11 satellite failed in Septem- ber 1994. TOVS ozone data were used from that time A 03 = aAZ500 + bAZm + cAT50 +d. (1) until the SBUV/2 data from NOAA-14 became available in May 1995. The regression coefficients can be thought of as the following:

5. Forecasting the ozone field 80, , 8 o 80, a~ —, b ~ 3—, c ~ —^ (2) The routinely processed total column ozone data obtained from the SBUV/2 instrument exist as two hemispheric 65 x 65 polar grids. The UVI processor where SOJSX is the change in ozone with respect to translates these fields to a 1° x 1° or 181 x 360 point the heights and temperature fields from two days ago equal latitude/longitude grid. This is done to ensure to yesterday. The forecast ozone differences are added equal weighting of all latitudes. This ozone field is to yesterday's ozone field, producing a forecast ozone for "yesterday," and a forecast of "tomorrow's" ozone field for tomorrow, as shown in Eq. 3. This proce- field must be made in order to create a forecast of dure, including the generation of new regression co- tomorrow's UV levels. It has been shown that the to- efficients, is performed each day: tal column ozone field correlates positively with the

50-hPa temperature field (T5Q) (e.g., Miller et al. 1979) and, to a lesser degree, negatively with the 100- and 03 (tomorrow) = 03 (yesterday) + A 03. (3)

500-hPa geopotential height fields (Z100 and Z500, respectively). We assume that the relationship of The above approach, as opposed to a direct ozone- changes in the ozone field to changes in the height to-meteorological parameter regression, tends to pre-

Bulletin of the American Meteorological Society 7 733

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC serve the maximum and minimum values of ozone, 1994 is lower than it would be using persistence for a 2- which have the greatest impact on surface UV radiation. (yesterday to tomorrow) through 5-day forecast. The Figure 4 shows the scatterplots for the forecast forecast and persistence errors, as well as their differ- Northern Hemisphere (NH) ozone field versus the ences, are greatest when the dynamic range of hemi- observed NH ozone field for dates at or near the equi- spheric ozone is largest. Therefore, one would expect noxes and solstices. The March date shows the largest the errors in the NH to be largest in the February- dynamic range of ozone values (230-454 DU), and the April period and smallest in the July-September pe- September date shows the smallest (246-338 DU). In riod. However, even when the forecast error is at its general, the high-ozone values occur at high latitudes, largest (-21 DU), the percentage error of total col- and the low-ozone values occur in the Tropics. Table umn ozone is still quite small [21 DU (error)/300DU 1 displays the correlation coefficients (r), the mean (normal ozone) = 7.0%]. difference, and the rms error between the forecast A decision tree, shown in Fig. 5, has been devel- and observed ozone fields. The small differences oped for use in the event of missing satellite ozone imply little, if any, bias between the two fields. The data. In the previous paragraph, it was shown that in rms errors are small when related to the mean ozone the absence of continuous SBUV/2 data, using the pre- amounts (9 DU/300 DU = 3%, 14 DU/300 DU vious day's regression coefficients provides a better = 4.67%). forecast than persistence. This holds true even for a Will all the above work produce a forecast of ozone 5-day forecast. After that time, or in the event of the better than that obtained using persistence (i.e., us- SBUV/2 instrument failing, a switch over to TOVS ing yesterday's ozone field as tomorrow's)? Table 2 ozone data occurs. In preparation for such occurrences, illustrates that for the NH the rms forecast error dur- the 80J8X regression coefficients are determined ing the period from 24 December 1993 to 5 February concurrently with the SBUV/2, using TOVS ozone data. As mentioned earlier, this decision tree has already been shown to work in the case of the failure of the NOAA-11 SBUV/2 instrument in September 1994. The UVI program switched its source of total ozone data to the NOAA-11 and later the NOAA-12 TOVS. When NOAA-14 SBUV/2 data became available opera- tionally in May 1995, the program auto- matically switched to using that data. Although, during the interim, there were ozone data available from the SBUV/2 instrument on boardNOAA-9, they were not available as an operational product.

6. Computation of the dear- sky UV dose rate at sea level

Once a forecasted ozone field is available, the next step is to apply a radiative transfer model [in this case one developed by J. Frederick of the University of Chicago (Frederick and Lubin 1988)] at each of the 1° x 1° grid points to deter- FIG. 4. Comparison of the Northern Hemisphere total ozone forecast vs that mine the spectral irradiances at each observed by the SBUV/2 instrument for (a) 22 March 1993, (b) 23 June 1992, (c) 23 September 1992, (d) 23 December 1992. The dynamic range is greatest in wavelength between 290 and 400 nm. March and smallest in September. Other inputs include a parameterized

734 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 1. Resulting statistics of comparisons of forecasts and TABLE 2. Mean rms errors of Northern Hemisphere ozone observations for four NH seasons. forecasts and persistence of 2-5 days in length between 24 December 1993 and 5 February 1994. Date r Diff (DU) Rms (DU) Number of Regression Persistence 22 March 1993 0.97 0.01 14.7 days forecast rms error (DU) rms error (DU)

23 June 1992 0.94 0.53 9.3 2 days 14.51 16.54

23 September 1992 0.43 0.53 9.3 3 days 17.59 19.56

23 December 1992 0.49 -0.43 11.3 4 days 19.02 21.12

5 days 21.86 22.65 value representing the atmospheric optical thickness, the UV albedo of the surface, the grid location's latitude, the day of year, and the solar time of day. The "minutes to burn" time. Various dermatological and time of day is held constant (solar noon), as are the ocular organizations and specialists were consulted UV albedo (5%), and the optical depth (0.2). The during the development of these categories. spectral irradiance values are weighted by the CIE As a result of the World Meteorological Organi- action spectrum. These weighted spectral irradiances zation (WMO) Meeting of Experts on UV indexes are then integrated between 290 and 400 nm to pro- held in July 1994, standards were set to guide the duce an erythemal irradiance (mW nr2) or dose rate format of UV indexes worldwide. One of these stan- for clear-sky conditions at sea level. This value is ap- dards is to express the UVI as the solar noon propriate for a single instant in time. It was our ini- erythemally weighted dose rate converted from units tial view that the public would find an expression of of mW nr2 to a unitless value. The WMO standard the amount of erythemal effective UV radiation over UV index unit is equal to 25 mW nr2. Hence, when a period of time or a dosage rather than as a dose rate the NWS converted from the dosage UVI to the more meaningful. To this end, the dose rate is inte- WMO standard dose rate UVI, the value for the same grated over the solar noon hour (1130-1230 local amount of UV radiation increased by 11.1%. This standard time). The diurnal cycle, as viewed in Fig. 2, change became effective on 18 April 1995. of the dose rate in the midlatitudes is such that the To run the radiative transfer model for each of the erythemal irradiance does not change more than 5% 181 x 360 grid points even on a Cray-90 takes several during the solar noon hour. Multiplying the dose rate minutes of CPU time. An alternate method uses a -2 by 3600 s results in a noon hour dosage (mJ m ) dif- three-dimensional look-up table with ozone, latitude, fering by not more than 2% from the value derived and day of year as the inputs. A comparison of dose from the integration of the actual erythemal irradiance rates derived from each method results in a correla- values. This means that the NWS dosage value is al- tion coefficient of 0.999 with a mean difference of ways a slight overestimate. The magnitude of the 0.017 mW m~2. Using the look-up table allows the 181 overestimation will increase slightly toward the Trop- x 360 grid points to be determined in a few seconds. ics and decrease slightly toward the poles. The resulting dosage in units of hJ nr2 was the UVI during its "experimental" phase between 28 June 1994 and 7. Elevation adjustments to the sea level 18 April 1995. clear-sky values The EPA has developed five exposure categories of the UVI along with guidelines of appropriate ac- To make the dose rate values realistic, the NWS tions for the public to protect themselves from over- reviewed which adjustments could be incorporated exposure to UV radiation. The categories, the UVI into the sea surface clear-sky value before the imple- within each category, and a list of simple protective mentation of the 1994 summer season UV program. actions are shown in Table 3. The categories were de- The most obvious and simplest modification would veloped taking into consideration the UVI, the vari- be to introduce an adjustment due to elevation gain. ous skin types of the public, and their associated As one rises above the sea level, the thickness of the

Bulletin of the American Meteorological Society 7 735

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC elude all differences between being in a UV index decision tree. valley and on a mountain top, such as lo- Get [D-l] ozone (SBUV/2 and TOVS) files from front-end computer cal changes due to air pollution, albedo, and viewing geometry. Since we are Get D, D+l Z500, Z100 and T50 MRF fields concerned with large-scale effects and Get [D-2] ozone files not local effects, we use an elevation adjustment derived from Frederick's Get [D-l] and [D-2] Z , Z , and T files 500 100 50 model. It is of the form shown in Eq. 4: Convert from polar to lat/long fields

Determine differences ([D-l] - [D-2]) for 03, Z, and T fields adj + «lZsfc Determine regression between 03 and Z, T fields for NH and SH (4) Z Determine differences ([D+l] - [D-l]) for Z and T fields + ^2 sfc

Use regressions to determine ([D+l] - [D-l]) differences for 03 where aQ = -0.04556, ax = 6.62033, a. = -0.23067, and Z f is in kilometers. Add 03 differences to [D-l] field to get [D+l] 03 field 2 ' sfc The adjustment is a 6.34% increase for Use radiative transfer LUT to get "CLEAR SKY" sea level dose rates the first kilometer, and this rate de- Determine hourly dosage adjusted for elevation at MOS cities creases for each additional kilometer gained. The elevation adjustments are Create bulletin for dissemination made at each grid point using the topog- Ozone data accessing decisions. raphy heights contained in the MRF model. IF SBUV/2 ozone field is not available for [D-l], THEN the regression coefficients for the last available preceding day is used (up to 4 days) to create the ozone forecast. 8. Discussion of annual IF the most recent SBUV/2 data is older than 4 days, variation of clear-sky UV THEN the most recent TOVS ozone data is used. indexes

IF the most recent TOVS data is older than 4 days, Using the above methodology, analy- THEN IF there is a SBUV/2 ozone field available in the last 4 days, THEN it is used assuming persistence. ses of the clear-sky UVI for both hemispheres can be made for any day of the IF there is no SBUV/2 field in the last 4 days, year, with latitudinal coverage depen- THEN IF there is a TOVS ozone field available in the last 4 days, dent upon available observations by the THEN it is used assuming persistence. SBUV/2 instrument. Figure 6a shows IF there is neither a SBUV/2 nor TOVS ozone field in the last 4 days, the 1979-86 zonal monthly mean total THEN the program ends. ozone as detected by SBUV on Nimbus- 7 from 75°S to 75°N. Note the high- FIG. 5. Decision tree of UV index processing and data accessing made prior to each run of the UV index ([D] is today, [D-l] is yesterday, and [D+l] is ozone (> 380 DU) amounts at latitudes tomorrow). poleward of 40°N from February to May. A complementary feature in the Southern Hemisphere (SH) begins to troposphere is lessened. Consequently, the amount of form in July but never reaches the high-ozone scattering also decreases, and incident UV radiation amounts found in the NH because of the formation increases. The adjustment due to elevation has been of the "ozone hole" in the South Polar regions in Sep- studied by Frederick (1993, personal communi- tember. In the Tropics an annual cycle is also notice- cation), Blumthaler et al. (1992), and Blumthaler able. An ozone minimum (03 < 245 DU) occurs in et al. (1994). Frederick cites an increase of 6% per the northern Tropics from December to February. An km using model calculations, while Blumthaler cites area of high ozone never really occurs in the Trop- a larger increase of 14%-18% per km from direct ob- ics, but rather the belt of ozone values less than servations. However, these direct observations in- 275 DU is thinnest from August to October.

736 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 3. EPA exposure catagories, the range of UVI in each catagory, and simple protective actions to take for conditions within each catagory.

Exposure UV catagory indexes Protective actions

Minimal 0, 1, and 2 Apply skin protection factor (SPF) 15+ sun screen.

Low 3 and 4 SPF 15+ sun screen and protective clothing such as a hat.

Moderate 5 and 6 SPF 15+ sun screen, protective clothing, and wear UVA+B absorbing sunglasses.

High 7, 8, and 9 SPF 15+ sun screen, protective clothing, sunglasses, and avoid being in the sun as much as possible between 1000 and 1600 LT.

Very high 10 SPF 15+ sun screen, protective clothing, sunglasses, and modify plans to be not in the sun at all between 1000 and 1600 LT.

Figure 6b shows the calculated zonal monthly Argentina and Chile. Figure 8a shows the SBUV/2 mean clear-sky UVI values (hJ nr2) for the ozone ozone analysis for 5 October 1992, when ozone-poor concentrations in Fig. 6a. Note that the annual cycle air passed over Tierra del Fuego. The resulting with greatest values occurs when and where the sun clear-sky UVI field is shown in Fig. 8b. Note the is highest in the sky. Note also that the clear-sky poleward ridging of the 3-, 4-, and especially the UVI values are not exactly symmetrical from NH 5-UVI contours. This illustrates the severity of summer to SH summer. At similar latitudes, the peak higher UV radiation values in locations where much values in the SH are higher than the NH peak values. lower values are the norm. In the Antarctic Ocean This is due to several factors. The ozone values are area, this can have a large impact on the aquatic life about 10 DU lower in the SH latitudes during the (Smith etal. 1992). summer months. This, combined with the earth being 3% closer to the sun during the SH summer, helps to create higher clear-sky UVI values during its sum- 9. Cloud adjustments to dear-sky values mer period than during the NH summer period. Figure 7a is an example of total ozone concen- In reality, the earth is not cloudless. Clouds gen- trations in the NH during midsummer, as measured erally attenuate UV radiation reaching the surface by by the SBUV/2 instrument on the NOAA-11 satellite. reflection, refraction, and absorption. Under certain Note that the field is quite variable, with ozone overcast conditions, the clear-sky UV radiation can "troughs" and "ridges" existing just as in pressure be attenuated up to 70%. However, in some instances, fields. Figure 7b shows the resulting elevation- clouds can actually enhance the clear-sky value by re- adjusted clear-sky UVI for the ozone concentrations flection from their sides (Madronich 1993). This ef- in Fig. 7a. Note how much the longitudinal variability fect, however, is short in duration and small in area has been dampened. In fact, there is more variability coverage. Under scattered or partly cloudy skies, the longitudinally due to the change in elevation than due passage of clouds will expose the surface to a clear- to changes in ozone. Latitudinally, the UVI contours sky condition some of the time. This presents a con- have their highest gradient in the midlatitudes. flict over which to present to the public: the maximum Occasionally, during times of large changes in possible clear-sky UVI or a "realistic" UVI attenu- ozone longitudinally, there are quite noticeable ated by the forecast cloud conditions over a period changes in the clear-sky UVI. Such a time occurs of time (i.e., minutes or hours). The decision was in the austral spring, when the SH polar vortex made to be as realistic as possible. Thus, it became becomes oblong in shape and the ozone hole extends imperative to be able to include a mechanism for equatorward well into the midlatitudes. This region adjusting the clear-sky UVI values due to cloud of very low ozone values can pass over southern conditions forecast for the solar noon hour. The

Bulletin of the American Meteorological Society 7 737

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC cast probabilities of clear (O-Vio cloud cover), scattered (2/io-5/io), broken (6/io-8/io), and overcast (9/io-10/io) cloud conditions (Erickson 1988). The MOS guidance is restricted to the contiguous United States and Alaska; the clear-sky UVI for Hawaii and Puerto Rico cannot be adjusted by this method.

10. Determination of adjustments due to cloud probabilities

A simple method for intro- ducing clouds into the UVI forecast is to correlate MOS cloud probabilities with the ratio of the measured UV amounts to the clear-sky forecast. So that the wide variety of U.S. climatic conditions are represented, this calculation must use as many observation sites as possible within the United States. We use 1992 data to develop the coefficients and 1993 data to test the results. Since data from a spectroradiometer can be weighted with the CIE weighting function, they are most easily compared to the forecast. However, the only spectroradiometer data for 1992 within the United States are available FIG. 6. (a) Monthly zonal total ozone (DU) averages from 75°S to 75°N from the from a single Brewer instrument NlMBUS-7 SBUV instrument, (b) Calculated monthly zonal clear-sky UVI (hJ nr2) operated by the EPA at Raleigh, averages from 75°S to 75°N, resulting from the ozone values in (a). North Carolina. Broadband radiometer data, however, are NWS/Office of Systems Development/Techniques available from the NOAA, which has assisted in the Development Laboratory has had very good success data collection from a network of Robertson-Berger in interpreting the MRF model outputs and statisti- (RB) 500 meters since 1975. However, conversion of cally deriving probabilities of occurrences of such RB meter counts to physical units is not fully under- parameters as precipitation, clouds, hours of sun- stood, and some sites have had errors within the ob- shine, and solar energy during 3-h intervals through- served data (i.e., a temperature dependence and out the model's forecast period (Glahn and Lowry receiver cosine error). As a result, the values reported 1972; Jensenius 1988). These products and others from these instruments are not correctly scaled. We comprise the Model Output Statistics (MOS) pack- make use of the method described by DeLuisi and age. We quantify cloud cover using the MOS fore- Harris (1983) to convert from RB meter counts into

738 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC physical units (mJ cm-2). An example of the result of this conversion for noontime observed values from the RB meter in Detroit, Michigan, is shown in Fig. 9a. This is one of seven sites that currently produce data labeled "reliable" by NOAA/Air Resources Laboratory, which maintains the instruments. Frederick et al. (1993) have made use of a RB meter and have created a methodology of adjusting the data so as to fit inside a clear-sky envelope, as determined by a radiative transfer model. Our variation of this method is to determine the RB clear-sky values for day i as the maximum observation within a 15-day envelope centered about day i. In so doing, 352 RB unsealed clear-sky observations are derived from the 1992 dataset. After removing the few obvious outli- ers, we linearly regress the "unsealed" RB values against the corresponding clear-sky values from the radiative transfer model. This produces a means of adjusting the unsealed RB observations to fit within, or very close to, the model's clear-sky envelope. Figure 9b shows the noontime RB data for Detroit after scaling. This methodology was applied to the 1992 data at each of the RB stations: Albuquerque, New Mexico; Concord, New Hampshire; Detroit, Michigan; El Paso, Texas; Minneapolis, Minnesota; Salt Lake City, Utah; and Seattle, Washington. The forecast MOS cloud probabilities for clear, scattered, and broken clouds at these same cities are regressed against the ratio of the noontime clear-sky dose rate and the scaled 1992 observations. No coefficient is determined for overcast skies since its probability is algebraically dependent upon the other three probabilities (i.e., the sum of the probabilities equals one). The regression produced the following constant, clear, scattered, and broken cloud coefficients (P, Ps, and Pb, respectively) with their respective 95% con- fidence limits: FIG. 7. (a) Northern Hemispheric analysis of total ozone (DU) • constant {Const) = 0.316 ± 0.172, observations for 1 July 1994 from the SBUV/2 instrument on • clear (a) = 0.676 ± 0.037, the NOAA-11 satellite. The black dot over the polar region shows • scattered (a) = 0.580 ± 0.033, and the latitudinal extent of satellite observations, (b) Northern Hemispheric analyses of the calculated clear-sky UVI (hJ m-2), • broken (ab) = 0.410 ± 0.077. with effects of elevation included for the 1 July 1994 ozone analysis in (a). The ensuing equation for determining the cloud attenuation factor (CAF) is 0.316 for 100% overcast. In an attempt to account for

CAF = Const + acPc + asPs + abPb. (5) the effects of haze in the regression, the surface dewpoint was included as a predictor along with the Thus, the regression produces CAFs of 0.992 for cloud probabilities. The regression statistics were the probability of 100% clear sky, 0.896 for 100% recalculated, but the correlation coefficients did not scattered clouds, 0.726 for 100% broken clouds, and increase significantly enough to warrant inclusion.

Bulletin of the American Meteorological Society 7 39

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC change from 1992 to 1993, we use the scaling coefficients determined at each of the RB instrument sites to rescale the 1993 data. The cloud regression coefficients determined using the 1992 data are used with the 1993 MOS cloud probability forecasts, producing CAFs for each day at each site. The correlation coefficient(r) for the CAFs from the MOS forecasts and the "scaled" RB observations for 1993 is 0.668, which we admit is not very good. The CAFs are used as one means of comparison, as this removes the ef- fect of the annual cycle imbedded in the data. A similar procedure using regressions with the "most likely" cloud category produces a correlation coefficient of only 0.576. When compared with scaled RB observations, the forecasted UVI is within one index unit 65% of the time and within two index units 89% of the time. Figure 10a shows the time series of UVI forecasts and the scaled RB observations at Detroit for the days of 9 June (day 160) through 17 Septem- ber (day 270) 1993. The day-to-day variation of the clear-sky UVI is solely due to changes in ozone over- laid on top of the solar annual cycle. This clear-sky value roughly corresponds to the maximum envelope of values in the observations. Cloud-induced features in the observations are generally found in the UVI forecasts. However, the regression approach inher- ently fails to capture the highest and lowest observation points. Figure 10b shows the UVI forecast CAFs versus the observed CAFs (observed dose rate/clear- sky forecast) for the same dates as in Fig. 10a. The limited range of the UVI forecasts is vividly illustrated. The observations shows a much larger range of CAF values. Here the overestimation at lower CAFs and underestimation at higher CAFs of the UVI forecasts is shown.

FIG. 8. (a) Southern Hemispheric analysis of total ozone 12. Verification: 1994 observations for 5 October 1992 from the SBUV/2 instrument on the NOAA-11 satellite. The black dot over the polar After the issuance of the UVI forecasts began in region shows the latitudinal extent of satellite observations, the summer of 1994 to 58 cities within the contigu- (b) Southern Hemispheric analyses of the calculated clear-sky UVI (hJ m~2), with effects of elevation included for the 5 October ous United States, Alaska, Hawaii, and Puerto Rico, 1992 ozone analysis in (a). an effort was made to enlarge the number of verification sites. As a result, data were obtained from a collection of governmental agencies, cancer institutes, 11. Verification: 1993 and private firms. Table 4 displays the site location, operating agency, or institute; the nearest city for The 1993 RB data for the above cities are used to which a UVI forecast could be made; and the type of test the validity of the cloud regression coefficients. instrument (broadband or spectral) used. Once the Assuming that the relationship between the unsealed data were received from these sites, they were RB observations and the clear-sky UVI did not checked for possible errors in recording and timing.

740 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC of these clouds. 3) The statistics drive results toward the mean. One additional feature is that using the UVI and solar noon observations may present a conservative estimate of the reli- ability of the UVI. Figure 12 shows a closer look at the Bos- ton daily UV observations for 9-19 July 1994. This time, all of the 15-min UV observations during the course of the day are FIG. 9. (a) Daily unsealed RB noontime UV observations (blocks) and the clear-sky UVI plotted. Note that on some days, 2 (hJ nr ) (solid line) for 1992 at Detroit, (b) Daily scaled RB noontime UV observations just before the peak time of so- (blocks) and the clear-sky UVI (hJ nr2) (solid line) for 1992 at Detroit. lar insolation, clouds develop and cause a decrease in the UV Because of errors found, some sites were not used observations. The peak times of these UV observations in the verification statistics. Some of the broadband are not during the solar noon hour but before or after. instruments use a conversion factor from electronic The forecast UVI seems to perform better against the watts measured to an erythemal dose rate. We used peak observed UV observations, as opposed to the 1-h the manufacture's suggested values. However, we noontime average. This may be understandable since will show later that the manufacture's values con- the MOS cloud probabilities are applicable for a 3-h sistently resulted in dose rates that behaved curi- period, not just the instant of solar noon. ously with respect to the forecasts and with spectral Table 5 shows the resulting statistics for Boston instruments. and the 20 other verification sites. This table shows Figure 11 shows the noontime observations by a the averages and the variabilities of the forecast and spectroradiometer operated by the EPA at Boston, observed UVIs at the various sites. It shows the wide Massachusetts, with the UVI forecast and the clear- range in mean UVIs observed and forecast from the sky forecast during the verification period of 9 June northernmost sites to the southernmost and highest- (day 160) through 17 September (day 270) 1994. Note again that the variability of the UVI is not as large as that of the actual observations. Most of the time the UVI forecast captures the essence of the observations. However, on cloudy days, the regression method results in UVI forecasts that are too high. This is a general tendency that we attribute to several factors. 1) MOS rarely goes to 100% overcast/clear. 2) We have no information on the radiative characteristics

FIG. 10. (a) Daily time series of scaled RB noontime UV observations (line with closed circle), UVI forecast from regression coefficients (line with star), and clear-sky UVI (thick line) for the period of days 160-270 (9 June to 17 September) of 1993 at Detroit, (b) Scatterplot of the UV observations and UVI forecasts from (a).

Bulletin of the American Meteorological Society 7 741

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 4. Listing of all UV observation sites as of 1994, the agency running the site, the site location (if not at the city airport), the verification city, its location, the altitude (m), and the instrument used.

Observation Site Verification City Elevation Agency site lat(N)/long(W) city Code lat(N)/long(W) (m) Instrument

USDA Douglas Lake, MI 45.55784.66° S. St. Marie, MI SSM 46.45784.37° 220 Yankee

Oxford, OH 39.53784.72° Dayton, OH DAY 39.90784.20° 305 UVB-1

Fort Collins, CO 40.807104.75° Cheyenne, WY CYS 41.107104.82° 1871

Bondville, IL 40.05788.37° Peoria, IL PIA 40.67789.68° 201

Davis, CA 38.537121.76° Sacramento, CA SAC 38.527121.50° 7

Howland, ME 45.21768.70° Bangor, ME BGR 44.80768.82° 58

Geneva, NY 42.88777.03° Rochester, NY ROC 43.12777.67° 169

Griffin, GA 33.17784.40° Atlanta, GA ATL 33.65784.43° 315

Pullman, WA 46.757117.18° Lewiston, ID LWS 46.387117.02° 438

NOAA Concord, NH Same as city lat/long Concord, NH CON 43.20771.50° 105 RB500

Detroit, MI Same as city lat/long Detroit, MI DTW 42.23783.33° 202

El Paso, TX Same as city lat/long El Paso, TX ELP 31807106.40° 1194

Minneapolis, MN Same as city lat/long Minneapolis, MN MSP 44.88793.22° 255

Salt Lake City, UT Same as city lat/long Salt Lake City, UT SLC 40.777111.97° 1288

Seattle, WA Same as city lat/long Seattle, WA SEA 47.757122.30° 137

altitude sites. The table also shows the CAFs of the more variable the observed cloud conditions are with forecast (CAFUVI) and observed (CAFQBS) UVIs. This respect to the forecast cloud conditions. Correlation provides information on the cloudiness at each site. coefficients of the individual site's CAF^Udco and The standard deviation of the CAFs shows how much CAFUVI range from a poor 0.242 at the NOAA's Al- buquerque site to respectable 0.812 at the U. S. De- partment of Agriculture's Douglas Lake, Michigan, site. Note that the very clear sites have the highest

mean CAFQBS, low standard deviations, and correlate the worst against the CAFUVI, even though the mean difference between the forecasts and the observations is not very large. The correlation coefficient only shows the linear association between the two CAF's. The mean and rms of the forecast and observed UVI differences are indicators of the forecast UVI's cor- rectness. Here we see that only 5 of the 21 sites have FIG. 11. Daily noontime spectroradiometer UV observations (line with closed circle), UVI forecast (line with star), and clear- mean differences larger than ±1.0 index units. How- sky UVI (thick line) for the period of days 160-270 (9 June to 17 ever, the rms of the differences shows a range of September) of 1994 at Boston. variation between 1 and 2 index units. Note also that

742 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 4. (Continued)

Observation Site Verification City Elevation Agency site Iat(N)/long(W) city Code lat(N)/long(W) (m) Instrument

NOAA Albuquerque, NM Same as city lat/long Albuquerque, NM ABQ 35.057106.62° 1620 Yankee

Boulder, CO 40.017105.15° Denver, CO DEN 39.757104.87° 1625 UVB-1

Bondville, IL 40.05788.37° Peoria, IL PIA 40.67789.68° 202

EPA Raleigh, NC Same as city lat/long Raleigh, NC RDU 35.97778.78° 134 Spectral

Atlanta, GA Same as city lat/long Atlanta, GA ATL 33.65784.43° 315 Brewer

Boston, MA Same as city lat/long Boston, MA BOS 42.37771.03° 9

Gaithersburg, MD Same as city lat/long Washington, DC DCA 38.85777.03° 20

Local: Corpus Cristi, TX Same as city lat/long Corpus Cristi, TX CRP 27.77797.50° 13 SL501

MD Houston, TX Same as city lat/long Houston, TX IAH 29.97795.35° 33 RB500

Anderson Austin, TX Same as city lat/long Austin, TX AUS 30.30797.35° 189 SL501

Midland, TX Same as city lat/long Midland, TX MAF 31.957102.18° 872 SL501

Biospherics San Diego, CA Same as city lat/long San Diego, CA SAN 32.737117.17° 9 SUV-100

Local San Diego, CA Same as city lat/long San Diego, CA SAN 32.737117.17° 9 SL501

Local Honolulu, HI Same as city lat/long Honolulu, HI HNL 21.357157.93° 5 SL501

seven of the eight sites using the Yankee UVB-1 in- served UV index falls into the high and very high (H strument have a suspicious negative mean difference. and VH) categories. For such cases, correct H and VH This consistent negative bias may be due to a con- forecasts (X) were made 517 times. High and very version factor that is too large, resulting in observa- high observations (F) occurred 105 times when the tions that are consistently close to or greater than the forecast exposure category was MLorM. High and clear-sky UVI. However, given the values as they are, very high forecasts (Z) were made 142 times when Fig. 13 shows a histogram of the differences between the observed category was MLorM. These numbers the forecast UVI and the noontime observations. The UVI forecasts are correct (Idiff I < 0.5) 31.8% of the time and are within 1 UVI unit 75.9% of the time and within 2 UVI units 91.5% of the time. This is consistent with the rms differences discussed above. Binning the forecasts and the observations into the EPA's exposure categories results in the values shown in Table 6. The computed "percent correct" value for this table is 60.7%. As Table 6 illustrates, the UVI tends to be conservative, occasionally overestimating on days with observations of "minimal," "low," or FIG. 12. Diurnal spectroradiometer UV observations (solid thin "moderate" (MLorM) but rarely underestimating on line), noontime observations (line with solid diamond), UVI days of "high" or "very high" observations. It is im- forecasts (stars), and clear-sky UVI (thick line) for Boston for days portant for the forecast to be correct when the ob- 190-201 (July 9-19) of 1994.

7 45 Bulletin of the American Meteorological Society

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 5. Statistics of UVI forecasts and observations at 21 sites within the contiguous United States, sorted by instrument and latitude.

Location: No. Mean UVI Mean UVI Mean CAF Mean CAF Site lat(N)/long(W) Period recordings forecast observation forecast observation

Boston 42.37771.03° 94165--94268 91 5.36 4.51 0.76 0.64

Gaithersburg 38.85777.03° 9418' 7--94268 73 5.48 4.47 0.75 0.61

Raleigh 35.97778.78° 94180--94243 52 6.17 6.21 0.71 0.72

Atlanta 33.65784.43° 94165--94258 77 6.47 5.05 0.73 0.57

San Diego 32.737117.17° 94158--94247 76 8.21 7.56 0.89 0.82

Minneapolis 48.88793.22° 94182--94289 103 4.30 3.65 0.71 0.60

Seattle 47.457122.30° 94182--94297 111 3.91 3.44 0.75 0.65

Concord 43.20771.50° 94182--94297 111 4.33 4.08 0.74 0.71

Detroit 42.23783.33° 94182--94297 111 4.54 4.04 0.74 0.67

El Paso 31.807106.40° 94182--94207 26 9.37 8.20 0.90 0.79

Douglas Lake 46.45784.37° 94216--94265 46 3.87 4.25 0.72 0.80

Pullman 46.387117.02° 94194-94265 69 5.43 6.31 0.87 1.01

Geneva 43.12777.67° 94217--94265 47 4.53 4.66 0.76 0.79

Bondville 40.67789.68° 94158--94243 83 6.09 6.41 0.76 0.80

Boulder 39.757104.87° 94182--94243 60 7.66 6.58 0.85 0.73

Davis 38.527121.50° 9418 1--94265 81 7.27 7.63 0.94 0.99

Albuquerque 35.057106.62° 94158--94243 84 8.92 10.00 0.89 1.00

Griffin 33.65784.44° 9423 1--94265 34 5.98 6.09 0.76 0.77

San Diego 32.737117.17° 94182--94250 67 7.94 7.79 0.87 0.86

Austin 30.30797.35° 94182--94243 61 7.95 8.36 0.83 0.87

Corpus Cristi 27.77797.50° 94182--94243 61 7.71 8.17 0.79 0.84

All 21 stations 94158--94297 1524 6.03 5.82 0.79 0.76

are used to compute the probability of detection occurrences that were correctly forecast, with 1.0 be- (PoD), as illustrated in Eq. 6: ing a correct forecast of all H and VH cases. The PoD for the above values is 0.831, indicating a high detection rate. This is comparable to, or better than, other PoD = ———. (6) X+Y v 7 types of severe weather forecasts that the NWS issues. For instance, in the case of forecasting severe local Here, PoD is the fraction of actual H and VH event storms, the national average PoD for 1993 was 0.70.

744 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 4. (Continued)

Std dev CAF Std dev CAF CAF correl UVI mean UVI rms UVI STD Site forecast observation coefficient^) difference difference difference Instrument

Boston 0.13 0.19 0.500 0.84 1.49 1.23 Spectal

Gaithersburg 0.13 0.20 0.625 1.02 1.57 1.19

Raleigh 0.10 0.22 0.470 -0.05 1.67 1.67

Atlanta 0.10 0.21 0.404 1.42 2.25 1.75

San Diego 0.06 0.17 0.493 0.65 1.61 1.47

Minneapolis 0.15 0.20 0.653 0.65 1.15 0.96 RB500

Seattle 0.14 0.17 0.635 0.47 0.89 0.75

Concord 0.14 0.24 0.515 0.25 1.22 1.19

Detroit 0.11 0.19 0.598 0.50 1.22 1.11

El Paso 0.04 0.21 0.532 1.18 2.34 2.02

Douglas Lake 0.13 0.30 0.812 -0.38 1.26 1.12 YES

Pullman 0.07 0.16 0.543 -0.88 1.22 0.84 UVB1

Geneva 0.13 0.28 0.506 -0.13 1.41 1.39

Bondville 0.10 0.22 0.573 -0.32 1.49 1.46

Boulder 0.06 0.23 0.425 1.08 2.26 1.99

Davis 0.05 0.12 0.266 -0.36 0.70 0.68

Albuquerque 0.05 0.09 0.242 -1.08 1.45 0.96

Griffin 0.12 0.20 0.462 -0.11 1.44 1.42

San Diego 0.12 0.21 0.303 0.25 1.91 1.89 SL501

Austin 0.07 0.19 0.476 -0.42 1.60 1.57

Corpus Cristi 0.06 0.13 0.350 -0.46 1.21 1.16

All 21 stations 0.13 0.24 0.621 0.22 1.55 1.51

Another statistic is the false alarm ratio (FAR). The more often an event is forecast and does not The FAR is the fraction of all H and VH event fore- occur, the higher the score. A score of 0.0 indicates casts that were observed as MLorM as illustrated in no false alarms. The H and VH values give a FAR of Eq. 7: 0.215, indicating a low occurrence of false alarms. Lastly, the Critical Success Index (CSI) is the ratio of correct H and VH forecasts to the number of H and FAR = (7) X + Z' VH events plus the number of incorrect H and VH

Bulletin of the American Meteorological Society 7 45

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC TABLE 6. Groupings of UVI forecasts and observations by exposure catagory.

Forecast

Exposure Minimal Low Moderate High Very high catagory 0,1, and 2 3 and 4 5 and 6 7,8, and 9 10+ Total

Minimal 62 61 36 11 0 170

Low 14 154 101 35 2 306

Observed Moderate 0 61 271 93 1 426

High 0 7 98 417 8 530

Very high 0 0 0 71 21 92

Total 76 283 506 627 32 1524

forecasts, or a 100% clear-sky probability. In the opposite sense, the MOS forecasts will rarely forecast high, nonclear probabilities when clear conditions are actually ob- CSI = - (8) served. This means that the UVI forecast will rarely x+r+z' be low when high values actually occur. Although the UVI forecast is strictly for the noontime hour, The best possible score is 1.0, and the CSI for the it has been shown that occasionally convective clouds H and VH values is 0.677. form before noon and diminish the observed UVI. In conclusion, the UVI forecasts tend to overesti- Within a few hours, these clouds dissipate, and the mate the UV radiation during cloudy days and under- observed UV is very close to that forecasted. It has estimate it during clear days. This characteristic is part been shown that some of the data used in the ver- of the nature of using a regression scheme to provide ification are under question, which may drastically the CAF. However, the latter condition may be a re- alter the results. However, taking the data as they are sult of the fact that the MOS forecasts rarely provide currently, the UVI forecasts are within 1 index unit nearly 76% of the time. For the most important conditions, when the UVI values encompass the high and very high exposure categories, the probability of detection is high, and the false alarm rate is low.

13. Future improvements

In its current state, the UVI serves well as an index to forewarn the public against overexposure to UV radiation. Scientifically, there are several aspects to the calculation of the UVI that can account for more realistic atmospheric and surface conditions. They include using an improved radiative transfer model that allows for multiple layers of clouds and includes

FIG. 13. Histogram of the (UVI forecast-UV observation) the cloud characteristics of cloud fraction and UV al- differences in whole UVI units for all 21 sites between July and bedo. This necessitates a switch from MOS cloud October 1994. probabilities to the MRF, or the regional "eta" model

746 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC (Black 1994) cloud amounts. The radiative transfer Their means of forecasting the ozone levels for to- model should work well with information derived morrow have been shown to be an improvement over from either of these numerical models. Another ad- the persistence method. A three-dimensional look-up vantage of a model-generated cloud product is that it table replicates the computations of the clear-sky ul- is a uniform grid. The MOS products are generated traviolet radiation levels at the surface for all latitudes. for each forecasting station (city) and are, therefore, Adjustments due to elevation and clouds compare very irregularly distributed over the United States. well against surface observations. Current forecasts Using either of these models will enlarge the forecast of the UVI are within 1 unit for 76% of the studied domain from the continental United States to at least cases and within 2 units for 91%. UVI forecasts do the Western Hemisphere. The disadvantage of the tend to overestimate UV values under cloudy condi- MRF cloud data is that the resolution will be 111 km, tions and underestimate UV values in very clear con- which may be too coarse. The eta model output reso- ditions. Future plans include improvements in the lution, 30 km, is much finer. forecasting of ozone and clouds, the incorporation of Another source of improvement may be the inclu- haze, and use of a more robust radiative transfer sion of ozone in the MRF global model as a passive model to make use of the better cloud data. As the tracer. This means that there will be no active chem- NWS complies with WMO standards for determina- istry within the model, but the model will advect the tion of the UVI, a 10% increase over the 1994 sum- ozone horizontally and vertically. To do this requires mer values can be expected. the input of ozone values at multiple levels in the atmosphere and the calculation of potential vorticity Acknowledgments. The authors would like to thank John as a mechanism of movement of the ozone. If this DeLuisi and Detlef Matt of the NOAA/Air Resources Laboratory; James Gibson and Dave Bigelow of USDA/Colorado State Uni- ozone product is an improvement over the current versity; John Rives of EPA/University of Georgia; Bill Barnard ozone forecast procedure, we would be able to re- of EPA/Atmospheric Research and Exposure Assessment Labo- move the generation of daily regressions between the ratory; Rocky Booth of Biospherical Instruments, Inc.; Karen observed ozone fields and temperature and height Watson of the MD Anderson Cancer Institute; and Ann Agro of fields, thus saving quite a bit of CPU time. Scripps Memorial Hospital for sharing their UV observations with us. Without these observations, the credibility of the UVI would Further research on the effects of haze and tropo- be greatly reduced. We would also like to thank Paul Polger of spheric pollutants on the makeup of the direct and NOAA/NWS for assisting with the verification statistics. John diffuse components of UV radiation need to be con- Frederick of the University of Chicago is thanked for sharing his ducted. It is difficult to adequately relate the meso- radiative transfer code with us, as well as sharing his expert scale characteristics of pollutants with the current knowledge of the nature of UV as it passes through the atmosphere. Jim Kerr, David Wardle, and Robert Saunders of Envi- weather observations. One possible exception is haze. ronment Canada are thanked for sharing their knowledge and ex- Although it is difficult to parameterize the effects of periences in the implementation and running of the Canadian UV haze due to the lack of observations at the surface and index. The development of the UV index was sponsored in part throughout the planetary boundary layer, haze should through Interagency Agreement DW13936025-01-1 with the U.S. be relatable to the absolute and dewpoint temperatures Environmental Protection Agency. in the planetary boundary layer. Lastly, the surface UV albedo certainly is not constant spatially and temporally. A dataset of cloud-free References surface UV albedos needs to be created. This can easily be done using the data from the Total Ozone Map- ACS, 1995: Cancer Facts and Figures. Publ. 95-375-No. ping Spectrometer instrument and a collocated cloud 5008.95, American Cancer Society, 29 pp. Black, T. L., 1994: The new NMC mesoscale Eta model: Descrip- identification scheme. tion and forecast examples. Wea. Forecasting, 9, 265-278. Blumthaler, M., and W. Ambach, 1988: Solar UV-B albedo of various surfaces. Photochem. Photobiol., 48, 85-88. 14. Summary , , and W. Rehwald, 1992: Solar UV-A and UV-B radiation fluxes at two alpine stations at different altitudes. The National Weather Service makes use of exist- Theor. Appl. Climatol., 46, 39-44. , A. R. Webb, G. Seckmeyer, A. F. Bais, M. Huber, and ing observations and technology to derive a forecast B. Mayer, 1994: Simultaneous spectroradiometry: A study of the potency of ultraviolet radiation at the surface of solar UV irradiance at two altitudes. Geophys. Res. Lett., during the peak hour, with the inclusion of clouds. 21, 2805-2808.

Bulletin of the American Meteorological Society 7 747

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC Colmar Brunton Research, 1992: National Omnibus Telephone Madronich, S., 1993, UV radiation in the natural and perturbed Survey of 1000 People. Cancer Society of New Zealand. atmosphere. UV-B Radiation and Ozone Depletion Effects Coulson, K. L., and D. W. Reynolds, 1971: The spectral reflec- on Humans, Animals, Plants, Micro-organisms, and Mate- tance of natural surfaces. J. Appl. Meteor., 10, 1285-1295. rials, Manfred Tevini, Ed., Lewis Publishers, 17-69. Decima Research, 1993: An investigation of Canadian's atti- McKenzie, R. L., W. A. Matthews, and P. V. Johnston, 1991: tudes related to Environment Canada's UV index. 6 pp. The relationship between erythemal UV and ozone, derived DeFabo, E. C., and F. P. Noonan, 1983: Mechanism of immune from spectral irradiance measurements. Geophys. Res. Lett., suppression by ultraviolet radiation in vivo. I. Evidence for 18, 2269-2272. the existence of a unique photoreceptor in skin and its role McKinlay, A. F., and B. L. Diffey, 1987: A reference spectrum in photoimmunology. J. Exper. Med., 158, 84-98. for ultraviolet induced erythema in human skin. Human DeLuisi, J. J., and J. M. Harris, 1983: A determination of the Exposure to Ultraviolet Radiation: Risks and Regulations, absolute radiant energy of a Robertson-Berger meter sun- W. F. Passchier and B. F. Bosnajakovic, Eds., Elsevier, burn unit. Atmos. Environ., 17, 751-758. 83-87. Erickson, M., 1988: LFM-based MOS cloud amount guidance Michelangeli, D. V., M. Allen, Y. L. Yung, R. L. Shia, D. Crisp, for the contiguous United States. Tech. Procedures Bull. Ser. and J. Eluszkiewicz, 1992: Enhancements of atmospheric 378, National Weather Service, Washington, DC, 7 pp. radiation by an aerosol layer. J. Geophys. Res., 97, 865-874. Frederick, J. E., and D. Lubin, 1988: The budget of biologically Miller, A. J., R. M. Nagatani, J. D. Laver, and B. Korty, 1979: active radiation in the earth-atmosphere system. J. Geophys. Utilization of 100 mb midlatitude height fields as an indica- Res., 93, 3825-3832. tor of sampling effects of total ozone variations. Mon. Wea. , A. E. Koob, A. D. Alberts, and E. C. Weatherhead, 1993: Rev., 107, 782-787. Empirical studies of tropospheric transmission in the ultra- Nagatani, R. M., J. D. Laver, and A. J. Miller, 1977: Backscat- violet: Broadband measurements. J. Appl. Meteor., 32, ter ultraviolet (BUV) ozone analysis system. NWS/NMC 1883-1892. Office Note 132, 15 pp. Glahn, H. R., and D. A. Lowry, 1972: The use of model output National Cancer Institute, 1973-1991, 1994: SEER Cancer Sta- statistics (MOS) in objective weather forecasting. J. Appl. tistics Review. NIH Publ. No. 94-2789. Meteor., 11, 1203-1211. Smith, R. C., and Coauthors, 1992: Ozone depletion: Ultravio- Jensenius, J. S., 1988: Objectively forecasting sunshine. Wea. let radiation and phytoplankton biology in Antarctic waters. Forecasting, 3, 5-17. Science, 255, 952-959. Kondratyev, K. Y., 1969: Radiation in the Atmosphere. Aca- Wilson, L. J., 1993: Canada's UV index—How It Is Computed demic Press. and Disseminated. Environment Canada, Atmospheric En- Liu, S. C., S. A. McKeen, and S. Madronich, 1991: Effects of vironment Service, 3 pp. Anthropogenic aerosols on biologically active ultraviolet radiation. Geophys. Res. Lett., 18, 2265-2268.

y \

748 Vol. 77, No. 4, April 1 996

Unauthenticated | Downloaded 10/11/21 11:14 AM UTC