Elizabeth C. Weatherhead, PhD University of Colorado at Boulder. April, 2005 Revised October, Report to the U.S. Environmental Protection Agency

Transcription

1 Task (e) Report, Contract 4D-888-WTSA: Submitted for APM 227 Report on Geographic and Seasonal Variability of Solar UV Radiation Affecting Human and Ecological Health Elizabeth C. Weatherhead, PhD University of Colorado at Boulder April, 2 Revised October, 26 Report to the U.S. Environmental Protection Agency Task (e) Compare the UV-B flux measurements with the predictions of the National Weather Service Ozone Watch Program, together with an analysis of the differences and potential causes for the discrepancies These results have been peer-reviewed through a process supervised by EPA. Cooperative Institute for Research in the Environmental Sciences University of Colorado at Boulder 32 Broadway Boulder, CO 83 betsy.weatherhead@colorado.edu +1 (33)

2 EXECUTIVE SUMMARY The U.S. Environmental Protection Agency s Ultraviolet (UV) Monitoring Program began collecting data using Brewer instruments at various national park and urban sites in 199. This report compares UV Index forecasts released by the National Weather Service with Brewer UV measurements at each of the sites. The National Weather Service and U.S. Environmental Protection Agency forecast a UV Index value for each day for several cities and locations across the U.S. The UV Index is computed according to criteria developed by the World Meteorological Organization (WMO). The prediction provides an estimate of the daily high UV irradiance for a particular location, and is based on estimates of total column ozone and cloudiness. Of 4,474 days of Brewer measurements, over 3,811 days are identified as good for comparison to the UV index. For each location, the daytime integrated values and the NWS UV Index forecasts are compared. The results are presented both in terms of correlation coefficients and absolute differences and indicate very good agreement. For mid and high-latitude locations, the correlations between the UV index forecasts and the daily measured UV levels are extremely high, ranging from.87 to.97. This extremely high agreement is impressive given the difficulty in predicting clouds and the fact that total column ozone measurements were missing for some periods of time, requiring climatological values to be used. For the low-latitude sites at Virgin Islands, Hawaii, Everglades, and Big Bend, the correlations are roughly between.6 and.8. This good agreement is surprising given that cloud estimates are not provided for the National Weather Service predictions of UV levels at two of the locations, and that clouds are difficult to predict in most locations. The differences between UV index values and measured UV values suggest a very strong seasonal dependence, with the highest differences generally observed in the summer. The standard deviation of the differences was generally around one UV Index value. These values are summarized for each site. In a few cases, the UV index values were observed to be systematically high and the mean deviation was also high (1.49 and 2.46) compared to the measurements. This bias occurred at the two locations Virgin Islands and Hawaii where the UV index predictions did not include cloud estimates and were therefore expected to be high relative to the actual measurements. Overall, the agreement between the NWS-predicted values and the Brewer measurements was surprisingly good. The results support the usefulness of the measurements in verifying and potentially improving the NWS UV Index forecasts. The high correlations demonstrate both the skill of the UV Index forecasts and the high quality of the Brewer measurements. 2

3 3

4 TABLE OF CONTENTS 1. Introduction and Background page 4 2. Description of National Weather Service UV Index page 3. Description of UV Dose Calculations page 7 4. Methods page 8 Quality Assurance page 8 Computation of NWS and Brewer UV Values page 14 Comparison of NWS forecasts and Brewer Measurements page 2. Detailed Site Analyses for All 21 Brewer Locations page Intercomparison of Results for All 21 Brewer Locations page Conclusion page 8. References page 7 4

5 1. Introduction and Background Task (e) compare the UV-B flux measurements with the predictions of the National Weather Service Ozone Watch Program, together with an analysis of the differences and discrepancies is one of several tasks defined by the U.S. Environmental Protection Agency (EPA) to analyze measurements from the network of ultraviolet radiation instruments operated by EPA and the University of Georgia (UGA) in collaboration with the National Park Service (NPS). Other tasks address different elements of UV-B exposure and effects, and the results of those tasks are summarized in the individual task reports. The tasks defined by the U.S. Environmental Protection Agency at the start of the project are as follows: (a) Determine the trends in UV-B flux at the individual Brewer sites, at groups of similar sites and/or across the network. (b) Analyze the factors affecting the observations and trends at each site (and across the network, as appropriate) including correlations with changes in the ozone column, changes in stratospheric ozone, changes in ground level ozone, changes attributable to other pollutants or atmospheric constituents, etc. (c) Analyze the major factors affecting the UV-B flux, including solar angle, latitude, elevation, cloud cover, pollution levels and composition, etc. (d) Analyze the direct versus indirect exposures to UV-B radiation and the factors affecting the ratio. (e) Compare the UV-B flux measurements with the predictions of the National Weather Service Ozone Watch Program, together with an analysis of the differences and potential causes for the discrepancies. (f) Compare the Brewer UV-B measurements with that of the Tropospheric Ultraviolet (TUV) Model, Total Ozone Mapping Spectrometer (TOMS) measurements, and provide an analysis of the differences of those measurements and the causes of those differences. (g) Compare the UV-B measurements with air pollution measurements to determine the effects of tropospheric air pollution on UV-B exposures. (h) Compare trends in UV-B flux measured by the network at mid-latitudes in the United States to United National Environmental Program (UNEP) data.

6 (i) Determine the effect of clouds/haze/aerosols on UV-B exposure. (j) Analyze the directional diffuse (cloudless) sky irradiance in the nm (UV- B and 32-4nm (UV-A) wavelength bands as a function of aerosol optical depth. (k) Analyze the directional diffuse sky irradiance in the 29-4nm (UV-B and UV- A) wavelength band as a function of cloud cover, cloud type, cloud depth. (l) Analyze the reflectance (spectral albedo) for key materials (snow, beach sand, concrete, asphalt and water) in the 29nm-4nm (UV-B and UV-A) wavelength band, as appropriate for the network measurement sites. (m) Analyze the bi-directional reflectance (forward-scattering and back-scattering) for key materials (snow, beach sand, concrete, asphalt and water) in the 29nm 4nm (UV-B and UV-A) wavelength band, as appropriate for the network. Task (e) compares the Brewer UV-B flux measurements with predictions from the National Weather Service s ozone/uv prediction program. The National Weather Service has used ozone values to produce daily forecasts of the UV Index (UVI) for major U.S. cities for the past ten years. This report will compare the UV-B flux values measured by the Brewer instruments with the archived UVI forecasts obtained by National Weather Service personnel. 2. Description of National Weather Service UV Index The National Weather Service and U.S. Environmental Protection Agency teamed up in 1994 to produce standardized UV Index forecasts for major cities across the United States. Canada had actually started producing UV Index forecasts in 1992, and similar products were being developed in many other countries. Two World Meteorological Organization Meetings of the Experts in 1994 and 1997 helped define the term UV Index (UVI) and develop standards for producing forecasts. A follow-up meeting by the World Health Organization in 21 further refined the UV Index definition in the context of health messages and exposure categories, and many countries have now switched over to this standard (WHO, 22). The U.S. adopted this new worldwide standard in 24 (Long, 23). The National Weather Service/U.S. Environmental Protection Agency UV Index forecasts are made for noontime of the following day. The archived forecast values can be compared with noontime measurements from the Brewer instruments to evaluate the accuracy of the forecasts. The UVI predictions are made for all major urban areas and 6

7 regions across the U.S., so the predictions can be compared with data collected at each of the 21 Brewer sites. The procedures for computing the UV Index are summarized by both the National Weather and the U.S. Environmental Protection Agency. In the U.S., the UV Index is computed using forecasted ozone levels, a computer model that relates ozone levels to UV incidence (incoming radiation level) on the ground, forecasted cloud amounts, and the elevation of the forecast cities. Other countries also incorporate ground observations into the forecast estimates. As reported in NWS and U.S. EPA documents (see, for instance, sunwise/uvcalc.html), the UV Index calculation starts with measurements of current total ozone amounts for the entire globe, obtained via two satellites operated by the National Oceanic and Atmospheric Administration (NOAA). These data are used to produce a forecast of ozone levels for the next day at various points around the country. A radiative transfer model is then used to determine the amount of UV radiation reaching the ground from 29 to 4 nm in wavelength (representing the full spectrum of UV wavelengths), using the time of day (solar noon), day of year, and latitude. This resulting irradiance estimates are weighted according to how human skin responds to each wavelength. The wavelength weighting is important because people need increased protection from wavelengths that harm skin than from wavelengths that do not damage people's skin. The weighting function is called the McKinlay-Diffey erythemal action spectrum, and is the action spectrum for human response adopted by the Commission Internationale de l'eclairage (CIE) (CIE, 1998). This weighting process is described in more detail below. The weighted irradiances are then integrated over the 29 to 4 nm range resulting in a value representing the total effect a given day's UV radiation will have on skin. In producing the UV Index, the weighted irradiance estimates are then adjusted for the effects of elevation and clouds. UV at the surface increases about 6 percent per kilometer above sea level. Clear skies allow 1 percent of the incoming UV radiation from the sun to reach the surface, whereas scattered clouds transmit 89 percent, broken clouds transmit 73 percent, and overcast conditions transmit only 31 percent. Once adjusted for elevation and clouds, this value is then scaled (divided) by a conversion factor of 2 and rounded to the nearest whole number. This process results in a UV Index that usually ranges from (where there is no sun light) to the mid teens (high UV). The computation of the UV Index does not currently include the effects of variable surface reflection (e.g., sand, water, or snow), atmospheric pollutants or haze. 7

8 3. Description of UV dose calculations A comparative biologically weighted UV value can be calculated from the Brewer UV irradiance measurements. UV amounts are expressed in terms of irradiance, which denotes the energy flow per unit area reaching a surface. Irradiance can be integrated over a spectrum of interest, for instance 29 to 4 nm for UV wavelengths reaching the terrestrial environment. To compute an effective UV dose rate and a time-integrated UV dose, the UV part of the spectrum is weighted with an erythemal response spectrum. An erythemal response spectrum demonstrates the human skin-reddening sensitivity to each UV wavelength and is determined using both field and laboratory studies. For human health effects, the action spectrum most commonly used is that of McKinlay and Diffey [12], given by 1, for 2 298nm; ( ) = a { 1., where a.94 (298. ), for nm; (1) b 1., where b.1 (139. ), for328 4nm. The McKinlay-Diffey curve represents the standard erythemal action spectrum adopted by the Commission Internationale de l'eclairage (CIE) to represent the average skin response over the UV-B and UV-A regions of the spectrum (CIE, 1998). Using the McKinlay-Diffey function, the erythemal UV irradiance, or dose rate, can be written as UV erythemaldoserate 4nm 2 F( ) ( ) d, (2) where F is the spectral irradiance, is the McKinlay-Diffey action spectrum, and the product is integrated over the entire wavelength range of interest. The resulting measurement is expressed as a dose rate in units of mw m -2 (WHO, 22). Integrating these values over the length of a day provides a daytime integrated UV dose. 8

9 4. Methods Denali Quality Assurance The first step in the analysis of the spectral UV data involved a review of the available data with the scientists at UGA to determine any potential biases at particular sites. UGA personnel have put considerable effort into producing a Level 1 data product suitable for scientific use. The Level 1 data from all 21 sites were used to address several questions about the various factors influencing UV. The spectral nature of the Brewer data, as well as the large geographic coverage of the Brewer network (Figure 1), allow for an analysis of changes in UV at various wavelengths, which can provide important information about the factors affecting UV throughout the continental United States and in Alaska and Hawaii. Denali Olympic Glacier Theodore Acadia Rockymtn Boulder Sequoia Canyonlands Albuquerque Riverside Chicago Gaithersburg Shenandoah Greatsmoky Rtp Atlanta Bigbend Everglades ii Hawaii Virginislands Figure 1. The map shows the locations of instruments for the EPA Brewer UV monitoring network operated by the University of Georgia. The sites cover a large geographical area and represent a range of ecosystems. Parameters relating to UV observations and temperature characterizations at each of the Brewer sites are summarized in Table 1. Columns a through c give the latitude, longitude, elevation, and year observations were begun at each site. 9

10 Table 1. Location information for the U.S. EPA Brewer instrument network. Site (Brewer #) (a) Latitude/ Longitude (b) Elevation (c) Start Date Glacier National Park, 48.7ºN, 113.4ºW 424 m 1997 Montana (96-134) Denali National Park, 63.7ºN, 149.ºW 661 m 1997 Alaska (96-141) Olympic National Park, 48.1ºN, 123.4ºW 32 m 1997 Washington (96-147) Rocky Mountain National 4.ºN, 1.ºW 2896 m 1998 Park, Colorado (96-146) Hawaii Volcanoes National 19.42ºN,.29ºW 1243 m 1999 Park, Hawaii Boulder, CO (93-11) 4.1ºN, 1.2ºW 1689 m 1996 Gaithersburg, MD (1) 39.1ºN, 77.2ºW 43 m 1994 Acadia National Park, 44.4ºN, 68.3ºW 137 m 1998 Maine (96-138)* Everglades National Park, 2.4ºN, 8.7ºW 18 m 1997 Florida (96-13) * Chicago, IL (94-13) 41.8ºN, 87.6ºW 16 m 1999 Atlanta, GA (94-18) * 33.8ºN, 84.4ºW 91 m 1994 Research Triangle Park, 3.9ºN, 78.9ºW 14 m 199 NC (92-87) Great Smoky National 3.6ºN, 83.8ºW 64 m 1996 Park, TN (96-132) * Big Bend National Park, 29.3ºN, 13.2ºW 329 m 1997 Texas (96-13) Albuquerque, NM (94-19) 3.1ºN, 16.6ºW 161 m 1998 * Sequoia National Park, 36.ºN, 118.8ºW 49 m 1998 California (96-139) * Virgin Islands National 18.3ºN, 64.8ºW 3 m 1998 Park, U.S. Virgin Islands (96-144) Shenandoah National Park, 38.ºN, 78.4ºW 32 m 1997 VA (96-137) Canyonlands National 38.ºN, 19.8ºW 814 m 1997 Park, Utah (96-133) Riverside, CA (94-112) 34.ºN, 117.3ºW 84 m 199 Theodore Roosevelt National Park, North Dakota (96-131) * 46.9ºN, 13.4ºW 238 m 1998 *A site that has a shift or other problem Problem values in the data, including missing scans or extremely high or low points, have been flagged in the Level 1 data files released by UGA. However, examination of the data has identified suspicious values that had not been flagged. The occurrence of such 1

11 values varies from site to site, with some sites achieving extremely good collection rates as well as high data quality. The results of extensive quality screening of the data are summarized in Figure 2. The data shown in red for each day have been evaluated as being problem-free, based on set criteria. Over 3, days of data have been identified as good: each of these good days contains between 1 and UV scans. 11

12 Acadia : 1693 good days FLAGS - IDENTIFIED AS GOOD DAYS Albuquerque : 168 good days Atlanta : 184 good days Bigbend : 142 good days Boulder : 18 good days Canyonlands : 1889 good days Chicago : 131 good days Denali : 112 good days Everglades : 1664 good days Gaithersburg : 239 good days Glacier : 1981 good days Great Smoky : 1797 good days Haw aii : 114 good days Olympic : 1899 good days R.T.P : 171 good days Riverside : 2328 good days Rocky Mountain : 1674 good days Sequoia : 1626 good days Shenandoah : 227 good days T. Roosevelt : 132 good days Virgin Islands : 136 good days Weatherhead Sun Sep 26 14:43:3 UMS 24 task1.ssc Figure 2. The data presented here are as released on the University of Georgia web site and screened for quality. These data are used in this report to examine spectral UV changes over the time periods of instrument operation. 12

13 The UGA staff screened the data based on the two criteria listed in the readme file associated with the site. The two criteria suggested for screening were that the data compared reasonably well to a clear sky model and that approximately the right number of scans were taken on a given day. In addition to the screening performed by UGA as discussed above, we employed at least seven additional criteria. These additional criteria eliminated roughly 12 percent of the daily values. The checks were necessary to prevent systematic problems which could have had large impacts on the results presented. However, detailed screening of the spectral data was not possible and it is likely that some critical problems still exist in the individual spectral values. The additional quality assurance tests are summarized as follows: Test 1: The data were checked to define the time of the first scan. For the daily data to pass this test, the first scan had to have occurred within one hour of when the first scan was scheduled to occur. Test 2: The data were checked to define the time of the last scan. For the daily data to pass this test, the last scan had to have occurred within one hour of when the last scan was scheduled to occur. Test 3: The minimum solar zenith angles were checked. This test verifies whether scans were taken near solar noon. For the daily data to pass this test, the minimum solar zenith angle recorded was required be within three degrees of the expected solar zenith angle for that time of year. Test 4: The number of scans in a single day was checked. This test evaluates whether enough data were taken to provide a reasonable estimate of the daytime-integrated value. The criterion for this test is that the number of scans must not differ from the scheduled number of scans for the time of year by more than 1. Test : The data were visually evaluated for spikes. Spikes are a well-known phenomenon with the Brewer instruments, though their causes are not completely understood. Spikes result in extremely large values for single wavelength measurement in a single scan and do not represent a robust measurement of atmospheric composition. Many spikes had already been removed through UGA s screening of the data. Test 6: The data were assessed for problems with inappropriate time sequences. Occasionally, the data were written to files incorrectly, with two days of data recorded to one daily file. In these cases, the daytime-integrated UV was observed to be much higher than usual. We identified and screened these instances by looking for cases in which the order of the solar zenith angles was out of line with what would happen during a normal day. The test also looked for uneven or highly irregular sampling during a day. 13

14 Test 7: The data were checked to assure that DUV values were within range. Observed daytime-integrated UV values are rarely above 7 oules/meter 2 /day and should never be equal to zero, given the sensitivity and locations of the Brewer instruments. All reported values greater than 9 and less than or equal to zero were omitted from further evaluation. No efforts were made to determine the cause of these unusual values, however an inspection of when they occurred at each site showed no patterns consistent with an atmospheric cause of the unusual values. Data were required to pass all seven tests to be used in the analysis of the Diffeyweighted UV. The number of data points removed in the screening process represented only a small portion of the data collected (less than 1% network wide), but the screening was necessary because only a few unusual points have the ability to change the results of the analyses. 14

15 Computation of NWS and Brewer UV values Figure 3 presents a graphical view of the number of days of Brewer data from the U.S. EPA network. For each site, the color-coding represents the number of days of quality measurements for a given year, displayed on the y-axis. Deep oranges and reds correspond to years with over 3 days of measurements available, while blues and greens indicate significantly fewer days of high quality observations. Number of Days Available EPA Brewer Data Acadia Albuquerque Atlanta Bigbend Boulder Canyonlands Chicago Denali Everglades Gaithersburg Glacier Great Smoky Hawaii Olympic R.T.P Riverside Rocky Mountain Sequoia Shenandoah T. Roosevelt Virgin Islands Figure 3. The color-coding denotes the number of days of high quality measurements available for each year at each of the U.S. EPA Brewer sites. Figure 4 summarizes the number of days with available NWS UV Index data. As in the above figure, reds and oranges correspond to a majority of values being available for the year, while greens denote years in which fewer (only 1-2) values were available. 1

16 Number of Days Available UV Index Data Acadia Albuquerque Atlanta Bigbend Boulder Canyonlands Chicago Denali Everglades Gaithersburg Glacier Great Smoky Hawaii Olympic R.T.P Riverside Rocky Mountain Sequoia Shenandoah T. Roosevelt Virgin Islands Figure 4. The number of days with available UV Index data is plotted for each year for each of the Brewer sites. The total column ozone time series plotted in Figure provides information about the ozone values used to calculate the NWS UV Index. For Atlanta, Georgia, in 1997, ozone peaked in the spring months and declined through the summer and into fall. The seasonal cycle is typical for ozone at midlatitudes. 16

17 Total Column Ozone F M A M A S O N D Atlanta, 1997 Figure. Total column ozone is plotted as a function of day of year for Atlanta, Georgia. For each site, the total column ozone is used in conjunction with a radiative transfer model to estimate the amount of UV radiation reaching the surface. The resulting clear sky UV Index forecasts based on the ozone values above are plotted in Figure 6. Cloud estimates are not used in this analysis, and the values for Atlanta in 1997 reflect a clear seasonal cycle: UV Index increases through the spring and reaches a peak in the summer months. This seasonal cycle is strongly related to Sun angle and is typical for most mid-latitude sites. 17

18 Clear Sky UV Index Forecast F M A M A S O N D Atlanta, 1997 Figure 6. The UV Index values plotted here for Atlanta are computed by the National Weather Service making use of only the Sun angle and the total column ozone values. The values in Figure 7 represent the cloud amounts used in the UV Index forecasts. The values shown here correspond to cloud conditions over Atlanta, Georgia, in Cloud conditions will vary for each of the U.S. EPA Brewer sites. 18

19 Cloud Amount Used in UV Index Forecast F M A M A S O N D Atlanta, 1997 Figure 7. The predicted cloud factors plotted for Atlanta, Georgia, are produced by the National Weather Service for the UV Index. A value of implies a strong influence of clouds, while higher values indicate higher transmission. The UV Index forecasts based on both cloudiness and ozone are plotted in Figure 8 for Atlanta, Georgia. The values indicate a larger spread than was present when only ozone and Sun angle were incorporated into the calculations. These UV Index values are compared with the UV values obtained from the Brewer instruments at each of the monitoring sites. 19

20 Actual UV Index Forecast F M A M A S O N D Atlanta, 1997 Figure 8. The UV Index forecasts released to the general public are plotted for Atlanta as a function of day of year. These values include the effects of Sun angle, total column ozone, and clouds, and are the values that are compared to the Brewer UV measurements. The UV values measured by the Brewer instrument at the Atlanta, Georgia, site are plotted in Figure 9. These values are computed using the biological weighting and integration over the wavelength spectrum as described in Section 1. As expected, the data reflect a similar seasonal cycle to those in Figure 8. These data from each measurement location are compared to the corresponding NWS UV Index forecasts for each site. 2

21 UV Measured Daytime Integrated UV F M A M A S O N D Atlanta, 1997 Figure 9. Actual, measured UV from the U.S. EPA Brewer instrument at Atlanta, Georgia, are plotted as a function of day of year. These values will be compared to the forecasted NWS UV Index values. Comparison of NWS forecasts and Brewer measurements Two methods are particularly useful for comparing measured UV with the NWS UV forecast. The first is a correlation of the two sets of daily values and the second is an estimate of the difference. The correlation allows us to identify whether the variations are similar in both datasets, but ignores offsets and changing slopes from location to location. The absolute differences is difficult can show results which magnify differences in high UV regions and times of year. Together, these two comparisons allow an effective comparison of the measurements with the UV index values. In order to estimate absolute differences, we make a conversion from daytime integrated UV levels, which are in oules/m 2 /day, to UV index units. For the entire network, we make a single multiplicative transformation based on comparison of all measurements at all sites. The factor we use is 1/6 to convert oules/m 2 /day CIE weighted UV to UV Index levels. Analysis of deriving individual factors at each site showed that this relationship was reasonable and would allow for an unbiased weighting from site to site. 21

22 4. Detailed Site Analyses For All 21 Brewer Locations The following pages of plots present the results for each Brewer site. It should be kept in mind that some of the results shown might not be entirely conclusive based on the short duration of the EPA Brewer datasets. For each of the sites, the differences between the NWS-forecasted UV Index values and weighted, daytime integrated UV measurements are plotted as a function of day of year. The differences are plotted in units of equivalent UV Index, so that a standard deviation of 1 corresponds to 1 value on the UV Index scale. 22

23 Acadia National Park Latitude: 44.4ºN Longitude: 68.3ºW Elevation: 137 m Standard Deviation: F M A M A S O N D Acadia Acadia National Park is located in the northeastern United States. The northern latitude location results in a very large seasonal cycle: average UV in the summer is more than eight times the average UV in the winter. Comparing the NWS values with the EPA measurements, we find that the agreement is generally quite good, and that the largest differences occur during the summer months. These months are when both the predicted and measured values are at their highest, and are also the months when clouds can play the greatest role in attenuating UV. 23

24 Albuquerque, NM Latitude: 3.1ºN Longitude: 16.6ºW Elevation: 161 m Standard Deviation: 1 - F M A M A S O N D Albuquerque The metropolitan area of Albuquerque is known for its dry, hot climate in the summer and generally mild winters. The southern location and high elevation of the site results in extremes of UV levels. The differences between the NWS values and the Brewer measurements are largest during the spring and summer, when clouds may have a large influence on the actual, measured UV. 24

25 Atlanta, GA Latitude: 33.8ºN Longitude: 84.4ºW Elevation: 91 m Standard Deviation: F M A M A S O N D Atlanta The twenty-county Atlanta metropolitan area is an important urban monitoring site for the Brewer network. The comparisons for this site indicate that the largest differences between the NWS and measured UV values occur in the months from March through October. Little difference is observed during the winter months when UV values are generally lower. 2

26 Big Bend National Park Latitude: 29.3ºN Longitude: 13.2ºW Elevation: 329 m Standard Deviation: F M A M A S O N D Bigbend Big Bend National Park covers over 8, acres in sourthern Texas and hosts an array of ecological, paleontological, and archaelogical research. The differences between the NWS and measured UV values can be large at all times of year at this site, and are both positive and negative, suggesting that neither value is systematically over- or under-estimating the UV amount relative to the other. The largest differences are observed during the spring, when ozone and cloud effects may play a large role. 26

27 Boulder, CO Latitude: 4.1ºN Longitude: 1.2ºW Elevation: 1689 m Standard Deviation: F M A M A S O N D Boulder Boulder s UV monitoring takes place within 3 miles of the city of Denver at elevation of 1689 meters. The forecasted and measured UV values agree fairly well during the late fall and through the winter. Larger differences are observed during the spring and summer. 27

28 Canyonlands National Park Latitude: 38.ºN Longitude: 19.8ºW Elevation: 814 m Standard Deviation: F M A M A S O N D Canyonlands Canyonlands National Park is located in the Utah high desert and is studied by numerous scientists because of its unique soil and a variety of algae, lichen and bacteria which allow for a rich desert ecosystem. At Canyonlands, the agreement between NWSforecasted and Brewer-measured UV is quite good during the late fall and through the winter months. During the spring and summer, the differences become larger but do not appear to be systemically biased toward either an over- or under-estimation of measured amounts. 28

29 Chicago, IL Latitude: 41.8ºN Longitude: 87.6ºW Elevation: 16 m Standard Deviation:.97 - F M A M A S O N D Chicago The nine-county metropolitan area of Chicago is located in the heartland of the United States and is home to eight million people. The NWS-forecasted and Brewer-measured UV values agree quite well in the late fall and through the winter. The differences start to increase in the spring and are largest at around the summer solstice. The differences do not seem to biased in either the positive or negative direction, suggesting that there is no systematic bias related to the treatment of clouds or other factors. 29

30 3

31 Denali National Park Latitude: 63.7ºN Longitude: 149.ºW Elevation: 661 m Standard Deviation:.79 - F M A M A S O N D Denali Denali National Park is the farthest north site in the EPA Brewer network. The data reflect a higher seasonality than those at any other site, with the winter values being generally quite low, often below the detection limit of the instrument. The differences between the NWS-forecasted and Brewer-measured UV are quite small, particularly during the fall and winter months. Even during the summer, when the largest differences are apparent, the values agree more closely than for other sites in the network. 31

32 Everglades National Park Latitude: 2.4ºN Longitude: 8.7ºW Elevation: 18 m Standard Deviation: F M A M A S O N D Everglades Everglades National Park is one of the more southerly sites in the EPA network. The summer UV levels are higher than the winter levels by only a factor of 2. The differences between the NWS-forecasted and Brewer-measured UV values are very large, particularly during the spring and summer months when clouds may play the greatest role. The differences are biased positively, suggesting that the forecast value tends to overestimate the measured UV. This difference is not unexpected, as the Everglades UV Index forecast values do not account for the effects of clouds. 32

33 Gaithersburg, MD Latitude: 39.1ºN Longitude: 77.2ºW Elevation: 43 m Standard Deviation: F M A M A S O N D Gaithersburg Gaithersburg, Maryland is located within 3 miles of both Baltimore, Maryland, and Washington, DC. The data show a large annual cycle and generally low variability. The differences between the two values are generally minimal in the late fall and through the winter, but increase during the spring and reach a peak around the time of the summer solstice. 33

34 Glacier National Park Latitude: 48.7ºN Longitude: 113.4ºW Elevation: 424 m Standard Deviation: F M A M A S O N D Glacier Glacier National Park encompasses over a million acres of land in the northern United States. The site experiences a large variation in the seasonal cycle of UV due to both its high latitude and high elevation (424 meters). The NWS-forecasted and Brewermeasured UV show little difference during the fall or winter. The largest differences occur during the summer months when the site experiences its largest UV values. 34

35 Great Smoky Mountains National Park Latitude: 3.6ºN Longitude: 83.8ºW Elevation: 64 m Standard Deviation: F M A M A S O N D Great Smoky Great Smoky National Park encompasses over a half million acres of land in the eastern U.S. The differences between predicted and measured UV at this site are largest during the spring and summer and likely due to changes in clouds and in the treatment of clouds in the UV forecasts. Agreement at this site during the fall and winter is not as tight as it is for other sites during these seasons, however the overall differences remain small. 3

36 Hawaii Volcanoes National Park Latitude: ºN Longitude: 1.28 ºW Elevation: 1243 m Standard Deviation: F M A M A S O N D Hawaii Hawaii Volcanoes National Park covers over 3, acres at a high elevation on a Pacific island. Hawaii and Virgin Islands are the only two locations for which the National Weather Service does not incorporate cloud predictions into their UV forecast. For this reason, the measurements are generally higher than than the forecasts resulting in the differences presented here being centered around 1., rather than around zero. 36

37 Olympic National Park Latitude: 48.1ºN Longitude: 123.4ºW Elevation: 32 m Standard Deviation:.8 - F M A M A S O N D Olympic Olympic National Park encompasses over 9, acres and is one of the most northernly parks in the network. As such, it experiences a strong seasonal cycle with summer measurements that are over 12 times higher than winter measurements. The differences between predicted and measured UV values at this site are minimal during the late fall and through the winter. The differences increase into the summer months, but are still not overly large. 37

38 Research Triangle Park Latitude: 3.9ºN Longitude: 78.9ºW Elevation: 14 m Standard Deviation:.93 - F M A M A S O N D R.T.P Research Triangle Park is one of the country s premier research areas in the southeastern United States. One of the first Brewer instruments was installed near RTP and the data record provides a long time series for comparing with NWS UV forecasts. The differences between the predicted and measured valus are largest during the spring, summer, and fall. The largest differences seem to result from the forecasted value overestimating the UV amounts, though there does not seem to be a systematic bias toward an overestimation of UV. 38

39 Riverside, CA Latitude: 34.ºN Longitude: 117.3ºW Elevation: 84 m Standard Deviation: 1. - F M A M A S O N D Riverside The Riverside-San Bernardino area is near the population area of Los Angeles which includes over 9 million people. The location is known both for its air pollution and its success in cleaning its air over the last decade. The differences between forecasted and measured UV are greatest during the late spring, but can be high anytime during the period from March to October. The forecasted UV seems to overestimate the measured UV more often than the reverse behavior occurs. 39

40 Rocky Mountain National Park Latitude: 4.ºN Longitude: 1.ºW Elevation: 2896 m Standard Deviation: F M A M A S O N D Rocky Mountain Rocky Mountain National Park sits on the east end of the Rocky Mountains. Its high elevation results in a very strong seasonal cycle, while its mid-latitude location results in generally low variability. The largest differences between predicted and measured UV occur during the late spring and summer, and may be associated with changes in cloud conditions, or in the treatment of clouds in the forecast model. The differences during the fall and winter are typically not as large. 4

41 Sequoia National Park Latitude: 36.ºN Longitude: 118.8ºW Elevation: 49 m Standard Deviation: F M A M A S O N D Sequoia Sequoia and King s Canyon National Parks cover over 8, acres in the mountains of California. At this site, the differences between the NWS-forecasted UV and the Brewermeasured UV are largest during the period from March through September. The larger differences are most often positive, suggesting that they may be caused by changes in cloudiness or by limitations in the cloud predictions used to estimate the UV Index. 41

42 Shenandoah National Park Latitude: 38.ºN Longitude: 78.4ºW Elevation: 32 m Standard Deviation: F M A M A S O N D Shenandoah Shenandoah National Park covers roughly 2, acres in the eastern U.S. The UV values over time indicate generally low variability. The differences between the predicted and measured UV are largest right around the solstice, but can be fairly large at almost all times of year. The biggest differences are positive, suggesting that the discrepancies between the predicted and measured values are most often due to an overestimation of the UV Index. 42

43 Theodore Roosevelt National Park Latitude: 46.9ºN Longitude: 13.4ºW Elevation: 238 m Standard Deviation: F M A M A S O N D T. Roosevelt Theodore Roosevelt National Park covers 7, acres in North Dakota. The Brewer UV measurements have shown little change over the past five years, and exhibit generally low variability. The differences between predicted and measured UV are largest during the late spring and through the summer and smallest during the fall and the winter. The differences are both positive and negative, suggesting no systematic bias between the two values. 43

44 Virgin Islands National Park Latitude: 18.3ºN Longitude: 64.8ºW Elevation: 3 m Standard Deviation: F M A M A S O N D Virgin Islands Virgin Islands National Park is the most southern location in the EPA netowrk. The Brewer is located within the tropics and therefore exhibits an unusual seasonal cycle, at least compared to what is typically observed at mid-latitudes. Virgin Islands and Hawaii are the only two locations for which the National Weather Service does not incorporate cloud estimates into their UV forecasts. For this reason, the differences of Brewer measurements minus NWS UV forecasts hovers above zero. 44

45 . Intercomparison of Results for All 21 Brewer Locations The analyses presented in this section summarize and compare the UV measured at the individual EPA Brewer sites with the predicted UV index values at the nearest city. There are several reasons why these comparisons should not result in perfect correlations. First of all, the sites are not co-located. Clouds are extremely difficult to predict, and changes in cloud conditions have been a substantial effect on the measured UV. Finally, the comparison is between the quality-assured daytime integrated values and the predictions, which are generally for local solar noon irradiances. Nonetheless, the results show that there is an extremely good relationship between the UV predicted for local noon and the UV measured for daytime integrals at each site. The correlation coefficients between the predicted and measured UV for each of sites are plotted in Figure 1. The correlation coefficients are generally extremely high, except at the lowest latitude sites, where they are still relatively high. The relationship shows that for the four most southern locations the correlations range between.7 and.8, while all of the other 17 sites show correlation coefficients much higher between.87 and.9. This high agreement is a strong testimony to both the quality of the UV index estimates and the quality of the EPA Brewer network. The reason for the lower correlation coefficients at the lower latitudes are the lack of good cloud predictions and the fact that ozone varies little at these locations. It should be noted that the National Weather Service incorporates clouds into their UV index forecasts for all locations except the Virgin Islands and Hawaii. The fact that clouds are not incorporated contributes to the lower correlations observed at these two locations. 4

46 Correlations.9 Albuquerque Canyonlands Olympic Riverside Sequoia T. Roosevelt R.T.P Chicago Acadia Atlanta Great Boulder Gaithersburg Smoky Glacier Shenandoah Rocky Mountain Denali.8.7 Bigbend Virgin Islands Everglades.6 Hawaii Latitude Figure 1. The correlation coefficients for the comparisons of measured UV at the Brewer sites and the NWS-forecasted UV index values are plotted as a function of latitude. The measured daytime integrated CIE levels from the Brewer network are plotted in Figure 11. At each of the sites, the seasonal cycle is clearly present, along with large day-to-day variations associated mainly with changing cloud conditions. The figure only shows the quality assured data from the network. These daytime integrated doses account for UV throughout the entire day, rather than being based on the UV radiation integrated near noon as is the case for the NWS UV Indices. The values are related, however, and both have biological relevance. 46

47 Acadia Albuquerque Atlanta Bigbend Boulder Canyonlands Chicago Denali Everglades Gaithersburg Glacier Great Smoky Haw aii Olympic R.T.P Riverside Rocky Mountain Sequoia Shenandoah T. Roosevelt Virgin Islands

48 Figure 11. The measured daytime integrated CIE levels are plotted for each day of each year for the 21 EPA Brewer locations. These data are compared with the NWS predicted UV Index values. The NWS-calculated UV Index values for clear sky conditions are shown in Figure 12 for each of the Brewer sites. These values are not released to the public because they do not include the effects of clouds. For clear sky conditions only, there is very little day-to-day variability in forecasted UV. The majority of the variation is due to seasonal cycle. Any other variations are a result of changes in total column ozone. These values are presented for comparison only and are not included in the analysis for each of the sites. 48

49 Acadia Albuquerque Atlanta Bigbend Boulder Canyonlands Chicago Denali Everglades Gaithersburg Glacier Great Smoky Haw aii Olympic R.T.P Riverside Rocky Mountain Sequoia Shenandoah T. Roosevelt Virgin Islands

50 Figure 12. UV index values calculated by the National Weather Service for clear sky conditions are plotted for each Brewer site. These values are not released to the public because they do not include the effects of clouds. They are also not the final UV Index values used in the comparisons with the Brewer observations. The UV Index values used in this analysis are those that are released to the general public. These values are plotted for each Brewer site in Figure 13. At each of the sites, the seasonal cycle is clearly evident. Large variations outside of the seasonal cycle are due primarily to changes in cloudiness. To a lesser extent, the UV Index values are influenced by changes in total column ozone.

51 Acadia Albuquerque Atlanta Bigbend Boulder Canyonlands Chicago Denali Everglades Gaithersburg Glacier Great Smoky Haw aii Olympic R.T.P Riverside Rocky Mountain Sequoia Shenandoah T. Roosevelt Virgin Islands

52 Figure 13. The UV index values released to the public, incorporating clouds as well as ozone changes, are plotted for each Brewer site. The seasonal cycle is clearly evident. Large variations beyond the seasonal cycle are due primarily to variations in expected cloudiness and to a lesser extent, variations in total column ozone. The differences between the NWS-forecasted UV Index and the weighted daytime integrated UV values measured by the Brewer instruments are shown in Figure 14. The results are reported in equivalent UV Index units, and the y-axis is held the same so that the differences at the sites can be compared. In general, the largest differences occur during the summer months when the UV values are largest. For all of the sites except Hawaii and the Virgin Islands, there seems to be no systematic bias with the UV Index either overestimating or underestimating the measured UV. At Hawaii and Virgin Islands, the UV Index forecasts do not incorporate clouds, hence at these two sites, a positive bias, with UV Index often overestimating the actual UV amounts, is observed. 2

53 Acadia 1.1 Albuquerque 1 Atlanta F M A M Bigbend A S O N D 1.79 F M A M A S O N D Boulder F M A M Chicago A S O N D.97 F M A M Denali A S O N D F M A M Gaithersburg A S O N D F M A M 1.14 Glacier A S O N D F M A M Haw aii A S O N D 2.4 F M A M Olympic A S O N D F M A M.8 R.T.P F M A M Riverside A S O N D F M A M A S O N D Rocky Mountain F M A M Shenandoah A S O N D 1.39 F M A M T. Roosevelt A S O N D A S O N D 1.17 A S O N D.93 A S O N D 1.2 Virgin Islands F M A M F M A M F M A M A S O N D Sequoia 1. A S O N D Great Smoky F M A M A S O N D Everglades F M A M F M A M Canyonlands 1.18 A S O N D 1.49 F M A M 3 A S O N D F M A M A S O N D

54 Figure 14. The differences between the National Weather Service UV Index forecasts and the measured, daytime integrated UV values are plotted as a function of time of year. The differences are reported in equivalent UV Index units. The measured daytime integrated UV values for each site are compared to the corresponding NWS UV Index values in Figure 1. The correlations are very strong, which is impressive given the difficulty estimating future UV levels and the difficulty in measuring UV accurately. The high correlations demonstrate the high quality of the measurements as well as the skill of the UV index forecasts. The comparisons of the differences between measured UV and UV Index values show the strength of the measurements in verifying and potentially improving the NWS UV Index forecasts 4

55 Acadia.91 Albuquerque Atlanta Bigbend Boulder Chicago Denali Gaithersburg 1 1 Haw aii 1 1 Riverside 1 1 Olympic Shenandoah T. Roosevelt 1 1 R.T.P Sequoia Virgin Islands Rocky Mountain 1 1 Great Smoky Everglades 1 Glacier Canyonlands

56 Figure 1. The measured daytime integrated UV values are compared to the National Weather Service s predicted UV levels. The strong correlations are extremely impressive given the difficulty of estimate future UV levels and the difficulty in measuring UV accurately. 6

57 7. Conclusion The National Weather Service makes daily predictions of UV levels for a number of U.S. cities. The value, referred to as the UV Index, is computed according to the criteria developed by the World Meteorological Organization (WMO). The prediction incorporates both total column ozone and cloudiness, and provides an estimate of daily high UV irradiance for a particular location. In comparing the UV index forecasts, we isolate the quality assured daily CIE-weighted UV data. Of the 4,474 days of data, over 3,811 are identified as good for comparison to the UV index. For each location, we compare the daytime integrated values and the NWS UV Index forecasts. The results are presented both in terms of correlation coefficients and absolute differences. For the low-latitude locations, Virgin Islands, Hawaii, Everglades and Big Bend, the correlations are high, roughly between.6 and.8. This high agreement is surprising given that cloud estimates are not provided for the National Weather Service predictions of UV levels at two of the locations (Virgin Islands and Hawaii) and clouds are difficult to predict in most locations. These high correlations serve as a testimony to both the National Weather Service s skill at predicting UV amounts and the quality of the EPA Brewer data. For mid and high-latitude locations, the correlations between the UV index forecasts and the daily measured UV levels are extremely high, ranging from.87 to.97. This extremely high agreement is extremely impressive given the difficulty in predicting clouds and the fact that total column ozone measurements were missing for some brief periods of time (several months), resulting in only climatological values being used. These extremely high correlations indicate the quality of both the NWS UV index forecasts and the EPA Brewer data. In addition to correlation, differences between UV index values and measured UV values were computed and shown to have a very high seasonal dependence, with the highest differences generally observed in the summer. The standard deviation of the differences was generally around one UV index value, with the exception of Hawaii, where the typical offset was 2.. In some cases, the UV index values were observed to be systematically high compared to the measurements. This bias occurred at the two locations where the UV index predictions did not include cloud estimates and were therefore expected to be high relative to the actual measurements. The comparisons of the differences between measured UV and UV Index values show the strength of the measurements in verifying and potentially improving the NWS UV 7