Draft Final Report CHARACTERIZING VISIBILITY TRENDS: A REVIEW OF HISTORICAL APPROACHES AND RECOMMENDATIONS FOR FUTURE ANALYSES SYSAPP-87/133 September 1987 EPA Contract No. 68-02-4373 Prepared for Office of Air Quality Planning and Standards U.S. Environmental Protection Agency Research Triangle Park, North Carolina Prepared by B. S. Austin L. R. Chinkin A. K. Pollack M. Moezzi C. S. Burton Systems Applications, Inc. 101 Lucas Valley Road San Rafael, CA 94903 And D. A. Latimer Gaia Associates Fort Cronkhite, Building 1055 Sausalito, CA 94965 ------- ACKNOWLEDGEMENTS The authors would like to express their appreciation to Neil Berg of the U.S. Environmental Protection Agency for his guidance and helpful sugges- tions during the project. We also wish to thank several individuals who provided their insight and knowledge about visibility. We thank Marc Pitchford of EPA Las Vegas, John Bachmann of OAQPS, William Wilson of ORD, William Malm of the National Park Service, John Trijonis of Sante Fe Research, Christine Sloane of General Motors, Scott Archer of the Colorado Bureau.of Land Management, Dennis Haddow of the U.S. Forest Service, Haluk Ozkaynak of Harvard University, Warren White'of Washington University, John Molenar of Air Resource Specialists, and Chuck McDade of Combustion Engineering. 87110 1 ii ------- EXECUTIVE SUMMARY The objective of this report is to review and recommend various approaches for analyzing and presenting visibility trends to be included as part of The U.S. Environmental Protection Agency (EPA) annual National Air Quality and Emissions Trends Report. Our review covers existing and potential data sources for visibility trends analysis, and parameters that have been or could be used to describe such trends. Our recommendations are based on an extensive review of published visibility trend studies, a telephone survey of individuals involved in developing visiblity trends in the United States, and a separate data analysis undertaken as part of this study. Specific areas addressed in the review and survey include alterna- tive physical parameters for describing visibility; alternative sources of data; statistical approaches for characterizing visibility frequency dis- tributions; and methods of accounting for natural visibility impairment. STUDY SUMMARY Visibility Indicators are significantly different from the traditional air quality indicators used in the Trends Report in that they report an air quality effect rather than air quality itself. This distinction is impor^ tant, both because air quality effects are not regulated under EPA stan- dards and because they are more difficult to measure and interpret than is air quality itself. In choosing an indicator for visibility trends, it is important to consider the relationship between the indicator and air quality. The two indicators that are used almost exclusively in visibility trends studies are the extinction coefficient (bext) and visual range. bext is a measure of light-absorbing and scattering particles in the air and is directly related to pollutant concentrations. Its disadvantages as an indicator are its inverse relationship with visibility and the lack of a long-term data base of direct measurements. It is also an unfamiliar technical term to readers of the trends report. Visual range is a measure of both the air quality effect of a given pollutant concentration and of unrelated factors such as the illumination of the atmosphere. It is not a direct measure of air quality. The advantages of using visual range as an indicator are that it is directly related to visibility, has a long-term iii 87110 1 ------- data base, and is easily understood by the general public. Using assumptions about the just-discernable contrast, bext can be related mathematically to visual range. Other indicators such as contrast, transmittance, or chromaticity may be appropriate for characterizing vistas of special importance, such as those in national parks, but cannot be used to characterize visibility in a concise and easy-to-understand way. There are only two potential data sources currently available for examin- ing long-term trends in visibility. The data base with the best spatial and temporal coverage is the U.S. Weather Service airport network, which has recorded hourly visual range observations at approximately 600 sites throughout the United States since the early 1940s. Use of this data for visibility trends analysis requires careful consideration of factors such as the availability of visibility markers past ten miles; inter-airport variations in reporting practices and marker availability; and changes in observers, locations, and marker types. Almost all trends studies use these data. The next largest data base is the National Park Service (NPS) monitoring network of 32 teleradiometer and camera sites in 28 western and three eastern national parks. Some of these sites have operated since 1978, but most have operated for shorter periods of time. To date, studies using these data have characterized current visibility conditions rather than examining long-term trends. Numerous other networks exist, but they have fewer sites and/or have operated for much shorter periods of time than has the NPS network. Two new networks that may be important for future visibility trends studies are the IMPROVE network and the Eastern Fine Particle and Visi- bility network (EFPVN). The IMPROVE network will assess progress toward - the national visibility goal and establish background levels of visibility and will consist of 20 transmissometer/camera sites in Class I areas. The EFPVN will collect the information necessary to determine the correct level of a secondary fine particle standard and will consist of 10 sites equipped with fine particle samplers, nephelometers, and cameras. The suitability of either of these networks for future trends analysis is questionable for several reasons: (1) Neither has guaranteed funding after an initial 3-4 year period; (2) neither has a large number of sites to adequately characterize visibility in the U.S.; and (3) with regard to the IMPROVE network, there are general uncertainties about transmissometer data quality. Visibility trends analysis has been performed using a variety of summary statistics such as means, frequency of visibility in certain classes, per- centiles, and ridits. Our discussion of these statistical methods centers on summaries of airport data since they constitute the most extensive data base available. Choosing an appropriate summary statistic for these data 87110 1 iv ------- is problematic because many airport visibility observations are underesti- mates of true visibility. For example, an observation of 20 miles may mean that the visibility marker at 20 miles can be seen and that the next marker (for example, a radio tower 30 miles away) cannot. The true visi- bility may be 25 miles and not 20. This factor makes it difficult to use means for visibility trends since the presence of such underestimates will lend a downward bias to the results. Another statistical technique, the flux method, reports the percentage of time that visibility falls within distance categories such as > 20 and < 30 miles. This method was used extensively in early visibility trends studies. Its advantage is that it summarizes exactly what the data report. However, it uses summary statistics that are unitless; thus actual trends in visual range across time are not reported as they would be by the means or median. The percentile approach reports trends in visibility equaled or exceeded some percentage of the time. (For example, the 90th percentile visibility would be the distance exceeded in only 10 percent of the observations) and is an extremely useful way of characterizing the entire visibility frequency distribution. Its principal disadvantage is that interpolation is often required to derive the percentiles of interest. Many current visibility trends studies use percentiles. Other approaches, such as ridit analysis, have been used primarily to validate other methods. SUMMARY OF RECOMMENDATIONS For the purpose of the Trends Report, it is desirable to distinguish visi- bility trends associated with anthropogenic factors from those associated primarily with natural factors such as rain, snow, fog, or windblown dust. A number of screening techniques have been used to eliminate such factors from the data. The least controversial is to remove hours with precipitation from the data base, since precipitation creates visibility impairment that is not related to pollutant concentrations. Removing fog is more complicated because it is not always easy to distinguish natural fog from anthropogenic haze. Since natural fog is unlikely when humidity is less than 80 to 90 percent, several techniques eliminate hours when humidity exceeds 80 percent. A problem associated with these techniques is that high humidities can affect the visibility reduction caused by a given concentration of particles; thus it may not be desirable to exclude these hours from the data. On the basis of our literature review and a prototype analysis (Section 6), we recommend removing hours with both fog and relative humidity greater than 85 percent from the data. We performed a detailed prototype analyis of data from St. Louis to pro- vide a further basis for our recommendations. In this analysis, we examined the selection of hours to be included in the data, the selection 87110 1 ------- of visibility categories for use in flux analysis, and the effect of tech- niques used for incorporating meteorological variability. It may not be desirable to use all daytime hours of data in the trends analysis because certain hours of the day (such as early morning) often experience natural visibility impairment not representative of visibility during the rest of the day. The choice of flux categories can also affect the estimated visibility trend. Our analysis indicates that the visibility trend was not appreciably altered by using data for only one hour per day (noontime), or by excluding observations for hours with precipitation or fog when humidity was greater than 85 percent. We also found that it is necessary to choose different flux categories for each individual station because there are large station-to-station differences both in reported values and reporting practices. Our recommendations for the current Trends Report are to use the NWS air- port data since no other data base approaches its geographic extent and long period of record. Although there are advantages and disadvantages associated with the use of any indicator, we recommend using visual range as an indicator because this indicator will make the visibility portion of the Trends Report useful to the largest number of people. At this time, we are recommending both flux and percentile statistical summaries, to determine whether both methods show the same general trends. If they do, we will recommend box plots of percentiles that are consistent with data displays currently presented in the Trends Report. We recommend examining trends in 10 cities chosen in conjunction with the EPA Project Officer for a 15-year time period. This time period is greater than that currently used in the Trends Report, but many visibility researchers believe that 5 or 10 years is not long enough to distinguish visibility trends from general meteorological variability. We do not recommend a longer period because of a major change in visibility reporting practices in 1970. We recommend using noontime observations and screening the data for precipitation and fog when relative humidity is greater than 85 percent, and when windblown dust is recorded. vi 87110 1 ------- CONTENTS Acknowledgments Executive Summary 1 INTRODUCTION 1 Study Background 1 Objectives of a Visibility Trend Analysis 3 Report Structure ' 4 2 ALTERNATIVE PARAMETERS FOR EVALUATING VISIBILITY TRENDS 5 Visibility Parameters 6 Relationships Among Visibility Parameters..... 7 Parameters Used in Trends Studies 9 3 SOURCES OF DATA FOR VISIBILITY TRENDS 14 United States Weather Service Airport Visibility Observations 14 Instrument-Based Visibility Data 21 The Improve Network 23 The Eastern Fine Particle Network 28 4 ALTERNATIVE APPROACHES FOR CHARACTERIZING TRENDS IN VISIBILITY 30 Describing Visibility Trends with Means 30 The Flux Method 31 The Percentile Approach 33 Ridit Analysis 38 vn 87110 1 ------- 5 ALTERNATIVE METHODOLOGIES FOR INCORPORATING METEOROLOGICAL VARIABILITY IN VISIBILITY TREND ANALYSIS 39 Selection of Hours 39 Use of Running Averages to Smooth Data 40 Elimination of Hours Subject to Natural Visibility Impairment 40 Treatment of Meteorological Variability by Other Researchers 41 6 PROTOTYPE ANALYSIS 43 Choice of Hours 43 Choice of Visibility Categories for Flux Analysis 45 Incorporation of Meteorological Variability 49 7 CONCLUSIONS AND RECOMMENDATIONS 53 Recommended Data Base 53 Recommended Visibility Indicator 53 Calculation and Display of Trend Statistics 54 Geographical Areas Recommended for Study 54 Recommended Temporal Features of Trend Analysis 55 Recommendations for Meteorological Screening Procedures... 57 Minimum Data Requirements 57 References 59 Appendix: SELECTED INSTRUMENT-BASED VISIBILITY AND FINE PARTICLE MONITORING NETWORKS 87110 1 vi ii ------- 1 INTRODUCTION The objective of this report 1s to review and recommend various approaches for analyzing and presenting visibility trends to be Included as part of the U.S. Environmental Protection Agency's annual National Air Quality and Emissions Trends Report. Our review covers existing and potential data sources used for visibility trends analysis and parameters that have been or could be used to describe trends. Our recommendations are based on a literature review, a telephone survey of individuals involved in develop- ing visibility trends in the United States, and a separate data analysis undertaken as part of this study. Specific areas- addressed in the review and survey include alternative physical parameters for describing visi- bility; alternative sources of data; statistical approaches for charac- terizing visibility frequency distributions; and methods of accounting for natural visibility impairment. STUDY BACKGROUND The U.S. Environmental Protection Agency (EPA) 1s considering adding an atmospheric visibility component to its annual National Air Quality and Emissions Trends Report, which is directed toward policy makers and the general public, and which contains summary statistics for atmospheric con- centrations of total suspended particulate matter, sulfur dioxide, carbon monoxide, nitrogen dioxide, ozone, and lead. The key objective of the trends report is to present statistics that describe current air quality, particularly as it relates to National Ambient Air Quality Standards (NAAQS) and trends during 5- or 10-year periods. The report compiles information from hundreds of monitoring sites into a highly condensed for- mat. Specifically, the annual trends report is currently designed to pro- vide information concerning (1) Trends in air pollutant concentrations (i.e., whether air pollu- tion 1s deteriorating, improving, or staying the same); (2) Whether NAAQS are being met or exceeded, and if exceeded, the average number of exceedances per year; 87110 8 ------- (3) Geographical variability in air quality; (4) Long-term average air quality as well as episodic, short-term averages for periods of high (peak) pollutant concentrations; (5) Disaggregated data for key urban areas and (6) Emission trends summaries used for comparison with air quality trends. The EPA's legislative responsibility for visibility is outlined in a num- ber of sections of the 1977 Clean Air Act Amendments, particularly sec- tions 109, 169A, and 165. Section 109 discusses air quality standards applicable to the entire United States; the latter two sections discuss issues applicable to pristine areas such as national parks. The inclusion of visibility in the Trends Report could function as an indicator of pro- gress toward achieving goals set forth in these sections. Section 109 requires the EPA to set primary (health-based) and secondary (welfare- based) standards for air pollutants. Visibility is considered to be an important welfare criteria. Since visibility impairment is primarily influenced by concentrations of fine particles (less than 2.5 ym) in the air, the EPA has recently initiated rulemaking for a secondary fine par- ticle standard based largely on fine particle visibility effects (Federal Register July 1, 1987). Section 169A establishes as a national goal "the prevention of any future and remedying of any existing, impairment of visibility in mandatory Class I federal areas where such impairment results from man-made air pollu- tion." Section 165 requires the states to include measures in their implementation plans (to achieve federal air quality requirements) to "prevent the significant deterioration" of air quality in areas with air quality better than the applicable national standard." These measures have historically taken the form of emission controls on new point sources that could degrade air quality in these areas over some agreed-upon increment. Most anthropogenic visibility impairment is a result of particles smaller than 2.5 micrometers in diameter (fine particles). Because of the rela- tionship between visibility and fine particle concentrations, visibility can serve as a surrogate for fine particles as well as an indicator of the visual effect of certain kinds of air pollution per se. In fact, visibil- ity is the public's primary basis for assessing pollutant levels. How- ever, natural factors such as rain, snow, sleet, fog, windblown dust, and forest fires also impair visibility. Furthermore, a given concentration of fine particles can cause different visibility effects, depending on the 87110 8 ------- relative humidity. Thus, care must be exercised in interpreting visi- bility trends to distinguish between trends in meteorological or climatic factors and trends in anthropogenic visibility impairment associated with fine particles. OBJECTIVES OF A VISIBILITY TREND ANALYSIS The selection of the methods to be used in EPA visibility trend analysis hinges on the specific applications served by the Trends Report. Visi- bility trends may be designed for the following specific objectives: PSD Class I visibility. The visibility trends report could be designed to relate specifically to the national goal expressed in Sections 165 and 169A of the Clean Air Act of preventing future and remedying existing visibility impairment in PSD Class I areas. Such trends reporting would be specific to those national parks and wil- derness areas (i.e., Class I areas) that are afforded visibility protection by the Act. Urban visibility. Since most Americans reside in urban areas, the visibility conditions found in these areas best characterize the typical, daily experience of the average American. Thus, urban area visibility could form the basis for exploring visibility trends. Rural (nonurban) visibility. If the objective is to characterize the visibility patterns of the land mass composing the United States, rural visibility would be most representative because urban areas and affected areas downwind of urban centers make up a relatively small .. fraction of the U.S. land area. A combination of these objectives. It may be appropriate to focus on a combination of these objectives. If so, it might be necessary to include a separate section in the Trends Report for each objective since there are fundamental differences in visibility, and in the importance of visibility, in each of these regions. For example, for reporting trends in Class I areas, the most interesting variable might be the number of days with exceptionally clear visibility, whereas in urban areas it might be the number of days with exception- ally bad visibility. Thus, the selection of specific data, indica- tors, and statistical parameters for visibility trends analysis and reporting should be based on a careful consideration of the specific uses of the visibility trends data. 87110 8 ------- REPORT STRUCTURE Our review of the various alternatives and components of visibility trends analysis is structured in the following manner: Section 2 discusses alternative parameters for evaluating visibility trends. Section 3 dis- cusses available and potential sources of data for visibility trends anal- ysis. Section 4 discusses alternative statistical methods for charac- terizing visibility trends. Section 5 reviews the various methodologies that could be used to treat meteorological variability in such analyses. A prototype analysis is presented in Section 6; Section 7 contains con- clusions and recommendations. 87110 8 ------- ALTERNATIVE PARAMETERS FOR EVALUATING VISIBILITY TRENDS EPA trends reports have traditionally presented air quality trends that are defined in terms of the composition of the atmosphere (e.g., Pb con- centration levels). In no case of which we are aware have trends in air quality effects (e.g., Pb levels in blood) been presented in these reports. This distinction is important for three reasons: (1) Tradi- tionally, it is air quality that is regulated through ambient standards that specify (among other things) permissable levels of atmospheric con- stituents after comprehensive consideration of their effects; (2).air quality can be measured to a desired degree of accuracy and precision but, because of numerous complicating factors, air quality effects cannot (for a cost comparable to that for measuring air quality); and (3) it is con- siderably less difficult to interpret air quality measurements than to interpret measurements of air quality effects. Visibility is an air quality effect. Its accurate measurement depends on numerous complicating factors involving how the atmosphere and, in turn, an object or scene, is illuminated and how the composition of the atmo- sphere varies with distance between the viewer and objects of interest. Interpretation of visibility measurements is made difficult by these fac- tors and also by gaps in our understanding of the factors influencing human visual perception. Visibility can be characterized by a number of physical parameters such as fine particle concentration, extinction, visual range, contrast transmittance, delta E, modulation transfer func- tion, and blue-red ratio. The fine particle concentration is an atmo- spheric constituent that is closely related to visibility whereas extinc- tion (or atmospheric transmittance for a sight path) is an air quality effect that is closely related to air quality; however, these relation- ships, which are well understood in principle, are difficult to treat in practice (Sloane and White, 1986; White, 1986). Parameters such as visual range, contrast, contrast transmittance, delta E, color contrast, modulation transfer function, or blue-red ratio depend on the illumination of the atmosphere and are poorly related to air quality; thus it is unlikely that they would form the basis of regulation, especially if traditional precepts in air quality management are fol- lowed. In this section, we discuss alternative physical parameters that 87110 9 ------- could be used as Indicators of visibility, their mathematical relation- ships, and the assumptions governing those relationships. A summary of parameters used in past trends studies is also included. VISIBILITY PARAMETERS The parameter most people are familiar with is visual range, i.e., how far we can see. Visual range has traditionally been defined in terms of the maximum distance from which a dark object can be seen against the horizon sky. However, when this definition is examined more closely, the actual visual experience is not so simple. In a valley just before sunset, for example, the visual effect of haze is quite different in different direc- tions. When looking toward the sun, the hills appear to be silhouettes with no detail; when looking away from the sun, all the rocks, trees, gul- lies, and other details on the hills are clearly discernible though the visual range is the same in both directions. This example suggests the importance of sun angle to visibility. The amount of detail that can be discerned on the hills corresponds to the contrast parameter. Another parameter, the extinction coefficient, can be derived from visual range and the "just-noticeable" contrast. This derivation requires a number of assumptions regarding the background sky radiance, and the uniformity of particulate concentrations in the atmosphere between the observer and the object. These three parameters—visual range, extinction, and contrast— are most often considered to be good visibility indicators. Visual Range (rv) The visual range, or "visibility," is the observation made at U.S. Weather Service monitoring sites in units of miles or kilometers. This parameter is one the public readily understands because of everyday experiences and its use in weather forecasts. Visual range is highest in conjunction with the lowest levels of air pollution, so it is inversely related to the atmospheric concentrations used in traditional trends reporting. Light Extinction Coefficient (bext) This parameter is a measure of the concentration of light-scattering and absorbing species in the atmosphere. Its units are inverse distance (e.g., m , km" , Mm ). The extinction coefficient is the equivalent light-scattering cross section (area) per unit volume of atmosphere, hence units of square meter per cubic meter or m^/m^ = m"*. Since it is similar 87110 9 ------- to a concentration, with increasing values corresponding to increasing pollution, its values can be directly compared with the composition of the atmosphere (e.g., fine particle concentrations). Light extinction is inversely related to visual range. Contrast (C) Contrast can also be used as a parameter. ' The contrast of a marker with the horizon sky is what the U.S. Weather Service observer responds to when making visual range observations. Contrast is also the parameter measured by a teleradiometer, from which other parameters can be calculated. How- ever, its value depends on the intrinsic contrast of the marker or target and the distance to it; thus, it is not a direct indicator of atmospheric conditions. It is specific to the particular line of sight used, unless a standard target and distance is used. Increasing contrast implies increasing visual range and decreasing extinction, and therefore decreas- ing atmospheric particulate concentration. Other Parameters Other much more sophisticated parameters have been used to characterize color and psychophysical conditions closely related to visual percep- tion. These parameters include transmittance, chromaticity, luminance, AE, modulation transfer function, and jnd. Their calculation requires very sophisticated and extensive measurements of the spectral properties of several lines of sight. Such parameters may be appropriate for charac- terizing vistas of special importance such as those in national parks, but they cannot be used to characterize visibility in a concise and easy-to- understand way. RELATIONSHIPS AMONG VISIBILITY PARAMETERS Visibility impairment can be defined as the loss of the contrast of objects viewed through the atmosphere. The contrast, C, of a landscape feature is the difference in light intensity of that feature when compared to its viewing background: where IQ and Ib are the spectral radiance (light intensities) of the land- scape object and the background, respectively. 87110 9 ------- The contrast of a landscape feature decreases as a function of the dis- tance between the feature and the observer in an exponential manner that can be described by the Lambert-Beer law: C(r) = C(0) exp(-bext r) . (2) where C(0) and C(r) are the contrasts of the landscape feature at distan- ces 0 and r (also called the intrinsic or inherent contrast and the apparent contrast, respectively), bgxt is the average light extinction coefficient of the atmosphere between the observer and the landscape fea- ture, and r is the distance between the observer and the landscape fea- ture. For black objects, C(0) = -1 and we can simplify Equation 2 as follows: C(r) = - exp(-bext r) . (3) The visual range distance (ry) is defined as the distance at which C(r) is the just-noticeable contrast or at the perceptibility threshold. If we define this contrast as Cmi-n, inserting these values into Equation 3 and solving for rv, we have the following equation (known as the Koschmieder equation): Traditionally, Cmin has been taken to be 0.02 (2 percent contrast) and has been used as the basis for the most widely quoted version of the Kosch- mieder equation: rv = -ln(0.02)/bext = 3.912/bext . (5) This commonly used formula for visual range (called the meteorological range) is based on the following assumptions (Middleton, 1952): (1) The background sky radiance at the observer is similar to that at the observed object location; (2) The distribution of the light-absorbing and -scattering con- stituents of the atmosphere is uniform over the line of sight between the object and the observer; and (3) The observed object is black (thus having a C(0) = -1); and (4) The threshold, or just-noticeable contrast, is 0.02. 87110 9 ------- The last two assumptions are rarely valid, whereas the first two may or may not be valid, depending on specific circumstances. Most viewed land- scape features are not black, and intrinsic contrasts for typical dark viewed objects range from -0.7 to -0.9 (Malm et a!., 1982). Field studies have suggested that a threshold contrast of 0.05 may be more appropriate for normal viewing situations. Inserting a value of 0.05 for Cm^n into equation (4) yields ry = -ln(0.05)/bext = 3.0/bext . (6) This formulation may be the most appropriate for relating U.S. Weather Service observations of visual range (rv) to light extinction coefficients (bext). However, other studies (Ozkaynak, 1985; Dzubay et. al, 1982) have suggested that Koschmieder constants in the range of 1.63 - 1.80 (instead of 3.9) might be most appropriate, at least for urban areas. This range would correspond to a contrast threshold of approximately 0.17. PARAMETERS USED IN TRENDS STUDIES Table 2-1 summarizes 30 visibility trend studies reviewed as part of this study. It is not a comprehensive list of all trends studies. The trends studies we reviewed all used either visual range or extinction coef- ficients to describe visibility. Different advantages and disadvantages are associated with the use of either parameter. Most people readily understand the meaning of visual range, but it is inversely related to atmospheric concentrations used elsewhere in the annual trends report. The terms extinction coefficient is less familiar, but can be thought of as a concentration of light-scattering and -absorbing species; it is directly related to fine particle concentrations. The reasons why trends studies do not use indicators other than visual range and/or extinction coefficients are straightforward. First, as will be discussed in Section 3, there is no long-term data. Collecting data for specialized parameters such as chromaticity or transmittance can be extremely expensive. Also, these parameters are not easily summarized. Which of these parameters should be used as a visibility indicator in a trends report? Tradition suggests that fine particle or extinction mea- surements would be preferable; fine particle concentrations are a measure of air quality, and extinction is a measure of an air quality effect that is closely related to air quality. A more intuitive measure such as visual range might be less preferable. However, as discussed in Section 3, there are no fine particle concentration or extinction records of suf- ficient scope (duration or spatial coverage) to use either of these 87110 9 ------- TABLE 2-1. Visibility trend studies. Reference Trljonls and Shapland (1979) Husar and Patterson (1984) Husar and Patterson (1986) Husar et al. (1979) Husar et al . (1981) (synopsis paper) Trljonls and Yuan (1978) Sloane (1980) Vlnzanl and Lamb (1986) Holzworth and Maya (I960) Miller (1972) Data Base 94 airports 137 airports 137 airports 70 airports 70 airports, ~ 30 turbidity monitors; 1 pyrhello- meter 100 airports; 6 sites from NASN parti cu- late network 15 airports 8 airports 2 airports 3 airports Time Period Analyzed 1974-1976 1948-1983 1948-1983 1948-1978 1960-1974 1916-1980 1974-1976 1948-1978 1949-1980 193S-1957 1962-1969 Area Entire Onited States Entire United States Entire United States Eastern United States Eastern United States Northeastern, eastern United States Mideastern United States Upper mid- western United States (Illinois region) Central and southern California i Ohio, Kentucky, and Tennessee Temporal Features 1n Analysis Annual and 3rd quarter analysis, noontime observations Seasonal analysis Noontime observa- tions, calendar quarterly analysis Seasonal analysis, noontime observations Dally, weekly, seasonal . 10 year analysis Annual , seasonal analysis; dally aver- age of four daylight observations Daylight observations at 10 a.m., 1 p.m., and 4 p.m. Seasonal analysis Seasonal and annual analysis Daytime observations in May, July, Septem- ber, and November Summer daylight observations Parameters Median midday visual range 50th percentlle bext 25th, 50th, 75th, 90th percentlle bext Median midday bext Mean bext, average turbidity coefficient 10th, 50th, 90th percen- tlle visual range 60th, 90th cumulative percent lies visibility In miles Visual range 60th, 90th percent lies, mean Mdits Percent frequencies of visibility between 0-10, 11-19," 20-30, > 30 miles Frequency of visual range < 6 miles and > 6 miles Metorologlcal Adjustments and Other No data eliminated RH, rain, snow, and fog adjustments Performed analysis with (1) no data el mi na ted (2) precipitation and fog eliminated (3) precipita- tion and fog eliminated, rh adjusted l%-2% of data discarded due to heavy rain, snow, or fog No data eliminated No data eliminated Rain and RH > 90% eliminated Performed analysis with (1) no data eliminated (2) precipitation and RH > 90% eliminated Data eliminated for hours with precipitation and/or RH > 90% Performed analysis with (1) no data eliminated (2) precipitation and RH > 90% eliminated continued ------- TABLE 2-1. Continued Reference Halm and Molenar (1984) Tombach et al . (1987) John Mulr Institute (1983) Malm et al . (1981) 4 4 Latlmer et al. (1978) Trljonls and Yuan (1979) Craig and Faulkenberry (1979) C-b 011 Shale Venture (1977) Roberts et al . (1975) Ursenbach et al. (1978) Time Period Data Base Analyzed 27 NPS sites 1978-1982 In western parks; tele- radiometers 11 sites; 1981-1982 teleradlometers, nephelometers 25 NPS Sites; 1978-1981 teleradlometers 13 NPS sites; 1978-1979 teleradlo- meters 16 airports 1948-1970 17 airports 1948-1976 3 airports 1950-1974 Photometers NA Photometers NA Photometers NA Area Western United States Western United States Western United States Southwestern United States Western United States Southwestern United States Oregon Northwestern Colorado Arizona Utah i Temporal Features 1n Analysis Dally and seasonal analysis Annual and seasonal analysis Monthly analysis Observations at 9 a.m., noon, 3 p.m.; seasonal averages Seasonal , annual analysis; daylight observations 3-year moving aver- ages; observations at 8 a.m., 11 a.m., 2 p.m., and 5 p.m. Daytime observations; seasonal analysis NA NA NA Parameters Median SVR calculated from radiance 10th, median, 90th percentlle, geometric mean, visual range Geometric mean, standard visual range Cumulative frequencies of standard visual range Percent of time visi- bility exceeds various values 10th, 50th, and 90th per- centlle visual range Mean rldlts 10th percentlle, median visual range 10th percentlle, median visual range Median visual range Meteorological Adjustments and Other Data validation procedures to determine accuracy of measure- ments, based on meteorological conditions and consistency checks Radiance measured at 405, 450, 550. and 630 ran for 5-6 vistas 1n each park Data for hours with fog and precipitation eliminated No data elimination mentioned Data for hours with precipita- tion or RH > 90% eliminated continued 87iiO 2 ------- TABLE 2-1. Concluded Reference Tombach and Chan (1977) Pueschel et al . (1978) Trljonls (1982) Duckworth and Klnney (1978, 1983) Nelburyer (1955) Espey, Huston, and Associates, Inc. (1984) Zlmmerroann and Wrlyht (1985) Kelso, Weiss, and Waggoner (1984) Roberts et al . (1974) Green and Battan (1967) Kelso (1981) Data Base Nephelometers Nephelometers, photometers 67 airports 1n' California; 19 studied A1 rport sites In California NA 16 airports 2 airports Nephelometers Photometers: black and white photos (densltometry) 2 airports Nephelometers Time Period Analyzed NA NA 1974-1976 1949-1976 1958-1977 NA 1971-1972 1976-1977 1981-1982 1960-1985 1979-1982 1973-1974 1949-1965 1975-1980 Area Utah Utah California California Los Angeles Texas Dallas and El Paso, Texas China Lake, California Petrified Forest, Arizona Arizona China Lake, i California Temporal Features In Analysis NA NA Observations at 1 p.m.; seasonal and annual analysis; 3-year moving averages Noontime observa- tions; summertime and annual analysis Annual and seasonal analysis; noontime observations Annual analysis; noontime observations Monthly average of 1-hour averages Annual and seasonal analysis Observations at 2 daytime hours Annual and monthly analysis Meteorological Parameters Adjustments and Other 10th percentlle, median visual range Median visual range 10th, median, 90th No data eliminated percentlle visual range Mean percent occurrence Data with RH > 70% eliminated of adverse noontime visibility over a 3- year period Median, 80th percentlle, mode. No data eliminated from noontime visual range 80th percentlle, days with Data for hours with fog and visual range > 15 and > 50 precipitation eliminated miles Mean bscat No data eliminated Mean, median, 95th percentlle; No data eliminated visual range Frequency of times per year No data eliminated visibility Is < 15 miles and < 30 miles Mean, median, 10th, 90th No data eliminated percentiles, bscat NA = not available. ------- parameters for trends analysis, whereas there are sufficient records of visual range. Visual range is also most closely related to people's everyday experience. Thus, for the objectives of the EPA trends report, visual range may be the best indicator of visibility trends. For these reasons, trends studies use visual range and/or extinction as indicators, often presenting graphical results with visual range on one side of the y-axis, and extinction on the other (for example, see Husar, 1979 and 1981). 87110 9 13 ------- SOURCES OF DATA FOR VISIBILITY TRENDS In this section, we discuss the data sources available for analysis of visibility trends. The two primary sources are U.S. Weather Service visi- bility observations made at hundreds of airports since the 1940s and the National Park Service (NPS) visibility monitoring network of approximately 31 stations operating since 1978. We describe these two networks in detail, focusing on their past and potential use in visibility trends analysis. Two other networks that may also be of importance in the future are the IMPROVE network and Eastern Fine Particle Visibility Network (EFPVN). We describe these networks at the end of this section. Repre- sentative long-term (over one year) visibility and fine particle monitor- ing networks in the United States are presented in Table 3-1. A brief discussion of selected monitoring networks that are limited in geographi- cal and/or temporal scope, but that have collected important data is presented in the Appendix. As Table 3-1 illustrates, the sources of visibility data available.for analyzing and reporting trends for the entire country are limited. The visibility data base with the best spatial and temporal coverage is the network of meteorological measurement sites operated by the U.S. Weather Service, primarily at airports. There are approximately 600 airport vis-- ibility measurement sites throughout the United States. Visual range and meteorological observations at these weather stations are made hourly. The National Park Service and other instrumented networks measuring visi- bility-related parameters provide much more limited spatial and temporal coverage; none of them covers the entire country and none has operated as long the ten years generally considered to be a rough minimum time period for a visibility trend analysis (Malm 1987a; Pitchford, 1987; Molenar, 1987; Middleton, 1987; Latimer, 1987). UNITED STATES WEATHER SERVICE AIRPORT VISIBILITY OBSERVATIONS Airport visual range estimates are made primarily for aircraft-related purposes and are therefore much more detailed and accurate in the 0-7 mile range than above 7 miles. For lower visibility readings, visibility obstructions such as precipitation, fog, or windblown dust must be reported. This focus on the lower range visibility results in several 87110 10 14 ------- TABLE 3-1. Visibility and fine particle monitoring in the United States. Study Name NWS Airport Data EPA-SAROAD System EPA IPMN NASN 'EPA-WPCS EPA-ORV 'EPA-RAPS IMPROVE Eastern Fine Particle Network NPS Regional Network NPS PMN SCENES EPRI-SURE Data Record 1948-present 1972-present 1978-1984 1956-1975 7/79-9/81 1980-1981 1975-1977 Mostly 1987 To begin fall, 1987 1978-present 1982-p resent 11/83-12/88 8/77-6/79 No. of Sites Measurements Approx. 600 Prevailing visibility TSP = ~ 2500 S02, N02, 03, S02 • ~ 530 TSP, CO, Pb 163 Fine, coarse, inhalable particles; sulfate 66 TSP, SOX 40 Fine, coarse particles; chem-analysis 3 Fine particles 20 Fine, coarse particles 20 Fine, coarse particles; radiance; visual range 10 bscat» visual range, fine particle mass 31 Radiance, visual range, fine particles Fine particles 6 Total , coarse, fine particles; aerosol chemical composition 56 S02, NO/NOX, 03; total, fine, Inhalable particles Instruments Human observation Various analyzers and samplers Filter media Particle samplers Particle samplers Filter media Particle samplers, nephelometers Cameras, teleradio- meters, fine particle samplers, transmissometers Fine particle samplers, nephelo- meters, and cameras Teleradiometers, cameras, particle samplers Particle samplers Teleradiometers, parti cul ate monitors, cameras Various analyzers and samplers Frequency of Measurements Hourly Varied 24-hr average, every 6 days Every 6 days 72 hr 24 hr every 6 days Every 3 days, continuously Hourly and 3 times per day 10 a.m. and 2 p.m. Camera--5-6 times daily; nephelometers continuously Every 3 days, continuously 24 hr Continuously and 4 times per day Area Urban and suburban, entin United States' Urban Many SAROAD sites Urban Urban Ohio River Valley Study St. Louis area Class I areas Eastern urban and suburban areas 28 western and 3 eastern Class I areas Eastern Class I areas Western United States 47 urban and 9 Class I sites continued 87ilO 2 15 ------- TABLE 3-1. Concluded Study Name EPRI-ERAQS EPR1-URAQS TVA-USMNP PANORAMAS RESOLVE Blue Hill Observatory Los Angeles Observations Data Record No. of Sites Measurements 6/79-12/79 9 S02, N0/N02. 03; total, fine. Inhalable particles 1/81-10/82 11 Fine, coarse particles; prevailing and viewing path visibility; radiance 1980-1981 1 Fine particles 7/84-10/84 26 bscat» f1ne« coarse particles 8/83-8/85 7 bext; radiance; fine, coarse particles 1889-1958 1 Visual range. "ext 1933-present 1 Visual range Instruments Telephotometers, nephelometers, cameras, particle samplers, human observation Teleradlometers, nephelometers. fine particle samplers, Human observation Particle samplers Nephelometers, particle samplers, cameras, tele- photometers Particle samplers, teleradiometers Human observation Human observation Frequency of Measurements Area Varied Nine "Class I site- Varied Nine western states Great Smoky Mountains National Par Daily Northwestern United State Varied California 8 a.m. and 2 p.m. Massachusett - Noontime Downtown Los Angeles 16 ------- limitations that must be considered when using these data for visibility trend analysis, and suggests that certain indicators and statistical descriptors of visibility would be more suitable than others. Trends studies based on airport data are used to illustrate our discussion of these limitations and their implications. The terms "visibility" and "visual range" are commonly used interchange- ably. Visual range is defined as the farthest distance at which a black object is discernible against the horizon background sky. Since the capa- bility of different observers to discern objects having small contrasts is likely to differ, visual range observations are subject to human per- ceptual and judgment differences and variations over time and are there- fore likely to be somewhat uncertain. By making assumptions regarding the "just-discernable" contrast and the nature of the specific line of sight, one can derive relationships between visual range and quantitative parameters that describe the atmosphere's ability to scatter and absorb light, as discussed Section 2. How Airport Visibility Measurements are Made Since black objects are rarely available at airports, visual range obser- vations are only approximations based on the "perceptibility" of avail- able, distant objects and are based on whether or not "markers" (e.g., buildings, towers, mountains) at different distances and in different directions are visible. There are standardized methods for observing and recording visual range at civilian and military airports in the United States. The U.S. Department of Commerce "Federal Meteorological Handbook No. 1, Surface Observations" documents the method by which visual range observations are made. When visual range is reported, standard operating procedures state that the criterion of "prevailing visibility" should be used. "Prevailing visibil- ity" is the "greatest visibility equaled or exceeded throughout at least half the horizon circle, which need not necessarily be continous." In determining prevailing visibility, the observer divides the horizon into sectors and assigns a visibility value to each. The number of sec- tors selected depends on the amount of variation in visibility between sectors. The observer uses well-known and documented (known distance or range) markers to determine visibility. However, over the course of sev- eral years, the location of visibility markers and even the airport itself may change, possibly introducing discontinuities into visibility trend data. 8711010 17 ------- If a visual range of a certain distance is recorded, that means the marker at that distance was visible, and that the marker at the next greatest distance was not visible. Thus, the reported visual range is probably understated, assuming a black target. After about 10 miles, the number of markers decreases rapidly and there are much greater distances between them. Before the 1970s, little effort was made to estimate visibility beyond 15 miles. If the 15-mile marker was very clear but a 30-mile mar- ker was not visible, a visibility of only 15 miles was reported. Visi- bility between.markers is currently estimated based on the clarity of the farthest marker that can be seen. Visibility past 15 miles is reported in five-mile increments. Data Availability Airport visual range observations are made every hour; however, in recent years, only the readings from every third hour are entered into the National Climatic Center's data base. Data are available in three differ- ent forms from the National Climatic Data Center in Asheville, North Caro- lina: (1) computerized records (TDF-14 format), (2) Local Climatological Data (LCD) summaries in hard copy, and (3) observer reports in hard copy or microfiche. While most of the data are available in both computerized and hard-copy format, data for some sites are available only in hard copy. Data Limitations There are a variety of limitations associated with airport visibility data. One is the general scarcity of markers beyond 10 miles, particu- larly for data obtained before 1970; another is interairport variation in reporting practices and marker availability. There can also be changes in observers, locations, and "type" of markers. The implications of these limitations are discussed in some detail below. Marker Availability At Individual Airports The reported visual range depends on the range within which the observa- tion occurs: within a 3-mile range, visual range is reported to fractions of a mile; between 3 and 15 miles, it is reported to the nearest mile; beyond 15 miles, the resolution is very coarse and somewhat arbitrary. Even in areas of the United States that have typically low visibilities, such as the East Coast, visibility is often greater than 15 miles. The availability of markers past 15 miles is crucial in determining trends in 8711010 lo ------- median and good visibility. However, most airports in the East and many airports in the West have few, if any, markers beyond 15 miles. Since 1970, airport observers have routinely estimated visibility between markers and beyond the farthest marker based on the clarity of the tar- get. This practice has resulted in a marked improvement in the availabil- ity of data for visual ranges greater than 15 miles. Many airports in the East have median visibilities around 15 miles, causing some researchers to report 60th percentile visibility to get around the pre-1970 truncation of data at 15 miles (Sloane, 1982a, 1982b). Many airports in the West have visibilities greater than 15 miles significantly more than 50 percent of the time. The frequency of visibility greater than 15 miles makes it difficult to analyze visibility trends using airport data from before 1970. Interairport Variations Variations in marker availability and reporting practices among airports have important implications both for choosing which airports to use in an analysis and for deciding how to describe typical visibility in a region. As another example, assume two airports have median visibilities of 75 miles. If at one of the airports the most distant marker is less than 75 miles the median would have to be extrapolated from the data and would probably be calculated as less than 75 miles. The airport with markers more distant than 75 miles will provide more data for calculating the median correctly. There are also important observer variations. - Some observers may care- fully estimate visibility between markers and beyond the farthest marker whereas others may not. In a survey of ten major airports conducted as part of this project, we found that the chief observer in Seattle almost never reported anything beyond the farthest marker observed. He inter- preted NWS observer regulations as meaning that if he could see the 50- mile marker but not the 100-mile marker, he should report 50. Other observers had a systematic method for reporting between markers.* Fin- ally, the "type" of marker can change. In one year many markers may be * For example, in New York there are no markers between 8 and 20 miles. If the 8-mile marker (the World Trade Center) is 'pretty clear', obser- vers will report 12 miles. If the observers can make out some of the colors in the banners on the center they report 15 miles (Thompson, 1987). 8711010 19 ------- tree-covered hills or mountains while in another many markers may instead be radio towers or buildings. The type of marker itself may have an effect on reported visibility since some markers are easier to see than others. If a single airport is used to describe visiblity for a given region, the presence of such reporting factors should be investigated and docu- mented. If more than one airport is used, attention must also be given to the fact that variations in visibility trends between airports may be obscured or exaggerated by differing reporting practices and marker availability. Problems inherent in using just one site can be more easily overcome than problems related to pooling sites. If data from different airports are pooled and averaged in some way, these considerations are even more important. Because of-the factors just discussed, regional-scale trends studies have generally carried out a telephone survey of potential airports before choosing which ones to use (Trijonis and Shapland, 1979; Trijonis and Yuan, 1978; Sloane, 1980; and Vinzani and Lamb, 1985). The survey makes it possible to find which airports have the best set of markers for visi- bility trends. Questions can focus on the number and types of markers at various distances, approaches for reporting visibility between markers and beyond the farthest marker; changes in personnel, and the effect of local pollutant sources on visibility. Some Implications Regarding the Use of Airport Data* In Visibility Trends Studies The nature of airport visibility data also has implications for the selec- tion of statistical descriptors of visibility parameters. For example, * NOTE: The U.S. Weather Service is in the process of evaluating alterna- tive instrumentation for measuring visual range (M. Pitchford, personal communication, 1987). If such a change in the manner in which the U.S. Weather Service measures visibility occurs, visibility trends analysis may be limited to two distinctly different time periods and measurement approaches: (1) The first time period, from 1948 to the transition time in the late 1980's or early 1990's, would be based on human observa- tions; (2) the second period, after the transition, would be based on the new instrumentation. It might not be possible to compare data before and after the transition because of differences in measurement technique. 87110 10 20 ------- since reported visibility is often understated, a statistic such as the mean can be a misleading indicator of true visibility conditions. Anal- ysis techniques need to be selected to minimize sources of uncertainty. For instance, analytic methods based on the cumulative frequency of visi- bility are considered to be best for analyzing airport data because they are well suited to the analysis of observations that are inherently under- estimates. Cumulative frequency and other methods are discussed more fully in Section 4. In spite of their limitations, the U.S. Weather Service visual range observations at airports appear to be the only currently available source of visibility data to support trend analysis and reporting for the entire country. Almost all trend analyses performed to date have been based on these data (Duckworth and Kinney, 1978,1980,1981; Husar et al., 1979, 1981, 1984, 1986; Latimer et al., 1978; Sloane, 1981, 1982a,b, 1983, 1984; Trijonis and Yuan, 1978, 1979; Trijonis, 1980, 1981, 1982, Trijonis and Shapland, 1979). Concerns have been raised about the accuracy and relia- bility of data based on human observations (Middleton, 1984); however, with careful screening of sites and observation periods, these data have been demonstrated to be useful for visibility trend analysis. INSTRUMENT-BASED VISIBILITY DATA Data obtained from instrument-based measurements of visibility or fine particles may be more precise than those based on human observations of visual range, and thus more reliable. However, each visibility measure- ment technique used currently is known to be subject to uncertainty. This uncertainty is related to a number of conditions that violate the assump-- tions of the methods used to estimate visual range from the parameter mea- sured (e.g., change in contrast, transmittance, and scattering coef- ficient). These conditions include bright and dark clouds behind the viewing target, bright haze, shadowed targets, and snow-covered targets (Malm, 1979; Allard and Tombach, 1981; Malm and Walther, 1980; Malm et al., 1982). The most extensive instrument-based visibility data base is the National Park Service (NFS) monitoring network begun in 1978 (Malm and Molenar, 1984). We describe this network and its use in trend and visi- bility characterization studies next. Other networks, including some particle monitoring networks, are described in Appendix A. All of these monitoring networks are summarized in Table 3-1. The National Park Service Visibility and Fine Particle Network The NPS has a responsibility to "conserve the scenery" (NPS Organic Act of 1916) and (along with other Federal Land Managers) to protect the air- quality-related values (Clean Air Act Amendments, 1977) of Class I areas 8711010 ------- under their jurisdiction. Visibility is considered to be an air quality- related value; thus, it is a critical aspect of the overall management of our national parks. Regulations implementing the Clean Air Act require visibility monitoring in Class I areas to determine existing visibility conditions, establish trends in visibility as emissions change, assess cause-effect relationships, review progress toward visibility goals, and provide input for modeling new sources. The NFS monitoring network is a result of this monitoring responsibility. The NFS visibility monitoring network was initiated in March of 1978 as a joint effort by the EPA Environmental Monitoring Systems Laboratory (EMSL) in Las Vegas and the NPS. In its early stages, it was called VIEW (Visi- bility Investigative Experiment in the West). Initially, 13 sites in west Texas, New Mexico, northern Arizona, western Colorado, and Utah were equipped with multi-wavelength teleradiometers and cameras. The teleradi- ometers were used to measure sky and.target radiance at different wave- lengths for approximately 5 targets in each of the 13 parks at 9 a.m., noon, and 3 p.m. daily. Contrast, standard visual range, and extinction coefficient were calculated from these measurements. Site operators also recorded meteorological conditions and target snow conditions. Associated with VIEW was a Western Fine Particle (WFP) program operated by the Uni- versity of California at Davis from late 1979 to late 1981. By 1982, there were 31 sites in the NPS network containing teleradiometers, cam- eras, and fine particulate samplers, three of which were in eastern Class I areas. Twenty of these contained both teleradiometers and particulate monitors. For a variety of reasons, only a limited number of trend assessments have attempted to make use of this data set. These reasons include (1) maximum amount of data available at any one site is 8 years, (2) this data is available from a small number of sites, and (3) technical problems with the teleradiometers make many of the measurements subject to uncer- tainty. In fact, data from this network has only been used to charac- terize current visibilty conditions and source effects. Malm et al. (1981) used the network to characterize visibility in the southwestern United States from the summer of 1978 to the spring of 1979, calculating standard visual range from the teleradiometer data. Malm and Molenar (1984) used this network to characterize western visibility from 1978 to 1982. Typical geometric mean visual ranges found in the summer of 1982 ranged from 40 to 144 km in California, Oregon, and Washington, and from 100 to 187 km in other parts of the West. These figures compare well with reported visual ranges obtained from airport observations by Trijonis (1979). Other studies have also used data from this network. Examples include Tombach et al. (1987), which examined annual and seasonal visibility from 87iio 10 22 ------- 1981-1982 at 11 sites; a study by the John Muir Institute, which operated the network for the first few years (John Muir Institute, 1982), that analyzed data for 25 sites for a three-year period covering 1978 through 1981, and a study that evaluated the impact of sources on visibility (Pitchford et al., 1981). IMPROVE NETWORK Description of the IMPROVE Network In 1984, the EPA reached a settlement agreement with the Environmental Defense Fund and the National Parks and Conservation Association over a suit filed because of the slow progress of states in responding to EPA visibility requirements (Stonefield and Metsa, 1987). The first part of the agreement required the EPA to propose and promulgate federal implemen- tation plans covering monitoring and new source review. The committee formed to oversee the monitoring efforts is called IMPROVE (Interagency Monitoring and Protected Visual Environments). IMPROVE is made up of a number of federal agencies, including the NPS, the EPA, the Forest Ser- vice, the Fish and Wildlife Service, and the Bureau of Land Management. IMPROVE's responsibility is to establish the background levels of visibil- ity necessary for assessing the impacts of potential new sources; to assess progress toward the national visibility goal; and to determine sources and levels of reasonably attributable impairment. The network set up to achieve these goals will consist of about 20 sites. All of the 156 United States visibility-protected areas were classified into four cate- gories on the basis of their potential for anticipated visibility change,' scenic sensitivity and value, existing visibility problems, and the number of other visibility-protected areas that could be represented. Sites in the first category had the highest monitoring priority. The locations of the monitoring sites are presented in Figure 3-1. Each site will use camera/transmissometer combinations and particulate samplers instead of teleradiometers. Some of the current NPS sites are included in the IMPROVE network. Current Status The IMPROVE network will not officially begin operation until fall of 1987. At present, less than five transmissometers are operating and bids are out to develop sites for the remainder of the network. The NPS is operating the network through a private contractor in Fort Collins, Colorado. 87110 10 23 ------- NPJS SCENEd f.ito.. — Auxiliary Sites — IMPROVE Network — New Auto Camera Sites Figure 3—la. Visibility monitoring sites. (See next page for key) ------- FIGURE 3-1 b. VISIBILITY MONITORING SITES AUXILIARY SITES 1. BUFFALO NR 2. CAPITAL REEF NP 3. CAPULIN MOUNTAIN NM 4. CHACO CULTURE NHP 5. COLORADO NM 6. CRATERS OF THE MOON NM 7. DEATH VALLEY NM 8. DINOSAUR NM 9. EVERGLADES NP 10. GRAND TETON NP 11. GUADALUPE MOUNTAINS NP 12. JOSHUA TREE NM 13. LAVA BEDS NM 14. LEHMAN CAVES NM 15. NORTH CASCADES NP 16. OLYMPIC NP 17. THEODORE ROOSEVELT NP 18. WIND CAVE NP 19. ZION NP 20. GLEN CANYON NRA 21. LAKE MEAD NRA NEW AUTOMATIC CAMERAS SITES 1. BADLANDS NATIONAL PARK 2. BANDELIER NATIONAL MONUMENT 3. GREAT SAND DUNES NATIONAL MONUMENT 4. HALEAKALA NATIONAL MONUMENT (NOT SHOWN) 5. LASSEN VOLCANIC NATIONAL PARK 6. POINT REYES NATIONAL SEASHORE 7. REDWOOD NATIONAL PARK 8. VIRGIN ISLANDS NATIONAL PARK ------- FIGURE 3-1 c. VISIBILITY MONITORING SITES IMPROVE NETWORK 1. ACADIA NP 2. BIG BEND NP 3. BRIDGER NF 4. BRYCE CANYON NP 5. CANYONLANDS NP 6. CHIRICAHUA NP 7. CRATER LAKE NP 8. DENALI NP (NOT SHOWN) 9. GLACIER NP 10. GRAND CANYON 11. GREAT SMOKY MOUNTAINS 12. JARBRIDGE NF 13. MESA VERDE NP 14. MOUNT RAINIER NP 15. ROCKY MOUNTAIN NP 16. SAN GORGONIO NF 17. SHENANDOAH NP 18. SUPERSTITION NF 19. WEMINUCHE NF 20. YOSEMITE NP 21. VOYAGEURS NP 22. ISLE ROYALE NP 23. HAWAII VOLCANOES NP (NOT SHOWN) 24. YELLOWSTONE NP 25. PETRIFIED FOREST NP 26. ARCHES NP 27. CARLSBAD CAVERNS NP 28. PINNACLES NP NhS SCENES SITES 1. -RYCE CANYON 2. GLEM CANYON 5. GRAND CANYON 4-. LAKE ------- Suitability for Future Trends Analysis The three key factors in identifying the suitability of the IMPROVE net- work for future trends studies are future funding, the number of sites, and the reliability of transmissometers. Funding for the network is cur- rently authorized by the EPA for the next three years. The federal land managers involved in the project are currently contributing 60 percent of the funding and may be willing, if budgets permit, to increase their involvement (Pitchford, 1987). In any case, future funding is uncer- tain. The small number of sites also poses some limitations for future trends studies in Class I areas. There are 156 Class I areas listed by the Department of Interior as having visibility important values. Although an effort was made to select 20 IMPROVE sites that were (among other things) representative of other sites, further assumptions of representativeness would need to be made if the 20 sites were used to estimate visibility trends in all 156 Class I areas. If we imagine the difficulty of charac- terizing visibility in a single large park such as Yellowstone with one monitoring site, we can begin to understand the difficulty of making such assumptions Malm (1987). It has been suggested that 30 to 40 sites in the West and 30 to 50 in the East might be enough to adequately characterize Class I area visibility. The transmissometer is still a relatively unproved instrument, which was only introduced into the field in 1986. This constitutes a limitation to the IMPROVE network because the quality of the data collected by the transmissometer is relatively unknown. Some testing and calibration will have to be performed to evaluate the performance of transmissometers in the IMPROVE network. The conversion to transmissometry as the principal method for long-path extinction measurements will require a number of tests to ensure the accuracy of the measurements: (1) Assessments of the comparability of transmissometer measurements with those collected by teleradiometers and other methods such as slide densitometers or nephelometers; (2) Comparison of measurements from two collocated transmissometers; (3) Introduction of filters into the line of sight to measure their effect on average line-of-sight extinction; and (4) Comparison of extreme values with theoretical maxima/minima. 87110 10 27 ------- EASTERN FINE PARTICLE VISIBILITY NETWORK (EFPVN) Description of the Eastern Fine Particle Visibility Network (EFPVN) The IMPROVE network will fulfill part of the ongoing effort to meet the national visibility goal. The Eastern Fine Particle and Visibility Moni- toring Network (EFPVN) is the result of recommendations made by the Inter- agency Visibility Task Force in 1985. The task force was charged with developing a long-term strategy for addressing visibility impairment from regional haze and integrating visibility issues with related regional issues such as acid deposition and fine particles (Bachmann, 1985). The overall purpose of the network is to facilitate the review of particulate matter air quality standards with respect to fine particles and visibility (Evans et al., 1987). The EPA is considering a fine particle standard that establishes specific visibility goals for regions in the East that are affected by regional haze, and has solicited public comment on the appropriateness of this focus (Federal Register, July 1, 1987). The EFPVN will collaborate with two other long-term networks in collecting data. The EVFPN will monitor scene-specific visibility, the optical pro- perties of the atmosphere, and aerosol characteristics, and will collect some meteorological data. There will be two types of sites: Tier 1 sites, which are primarily non-urban and will constitute the core research sites, and Tier 2 sites, which will gather more limited information. The 10 Tier-1 sites will use an automated fine particle sampler, a nephelometer, and an automated camera. Meteorological data collected will include wind speed, wind direction, temperature, change in temperature, relative hum- idity, and solar radiation. The Tier-1 sites will be located in each of ten areas identified by Husar (1984, 1986) as having similar visibility trends based on airport data (Evans, 1987). The 20 Tier-2 sites will be camera-only sites and will likely be located along with with existing SLAMS PM-10 monitors (Evans, et al, 1987). Current Status A test site at Whiteface mountain in New York is operating now with a nephelometer, quartz and teflon filter fine particle samplers, and two cameras (one half-way between the target and the first camera). The net- work should begin operation in October or November of 1987. As currently planned, it appears that only the Tier-1 sites will be included in the network because of funding limitations. 28 87110 10 ------- Suitability For Future Trends Analysis As with the IMPROVE network, a primary problem with the EFPVN is fund- ing. This network has been established largely to aid in determining the correct level of a possible fine particle standard (Pitchford, 1987). It has been funded for only four years, which is not considered to be an adequate time period for a trends study. 29 87110 10 ------- ALTERNATIVE APPROACHES FOR CHARACTERIZING TRENDS IN VISIBILITY Visibility trends analysis has been performed using a variety of different summary statistics, including means, medians, percentiles, frequencies of visibility greater or less than a reference distance, and ridits (proba- bility of a given visibility frequency distribution being better than a reference distribution). However, visibility trends analysis can be com- plicated by diurnal and seasonal variation. Diurnal variation has been addressed through examination of summary statistics for certain times of day only (most commonly midday) or several times of day excluding morn- ings. It is common to exclude nighttime visibility because it is not relevant to human visual perception. Seasonal variation has been addres- sed through the separate calculation of trends for each season. In this section, we discuss several approaches for summarizing and comparing visi- bility frequency distributions in an abbreviated, easy-to-understand fashion, and the implications and limitations associated with each approach. These approaches are reviewed here in the context of summari- zing airport visibility data. Our recommended approach is discussed in Section 7. DESCRIBING VISIBILITY TRENDS WITH MEANS It is difficult to summarize visibility data as reported by airport weather stations using means because of the nature of these data. Calcu- lating a mean value from airport data has been compared to "having one foot in a pail of boiling water and another in a pail of freezing water and saying that, on average, the temperature is luke-warm" (Sloane, 1987). As discussed in Section 2, each reported value is actually a lower bound to the actual visibility, particularly at longer distances. For example, if there are markers at 7 and 15 miles and the 7-mile marker is visible but the 15 mile marker is not, 7 miles might be the reported visi- bility, though visibility is actually between 7 and 15 miles. Thus, not only is calculation of the mean value an inappropriate method for sum- marizing data reported in such a manner, but it will also systematically underestimate true average visibility. None of the studies reviewed calculated simple arithmetic means for air- port data, though a few (Kelso, 1981; Kelso et al., 1984; Roberts, 1974) 87110 ------- have used this approach to summarize trends in visibility data collected with instrument networks. Other studies (Kelso et al., 1981; Husar, 1981; John Muir Institute, 1983; Roberts, 1974; Tombach et al., 1987) have used geometric means (means calculated from logarithms of the data), which are better measures of central tendency for the skewed distributions that might be observed in air quality or visibility data. For most of these studies, the means are calculated and compared with other measures of central tendency, such as the median, but are not normally used as the sole measure of "average" visibility. THE FLUX METHOD One way of presenting visibility trends is known as the flux method, which graphically portrays exactly the reported data. The trend parameters are the percentage of time that visibility is less than or greater than the discrete reported values. This method is well-suited to airport visi- bility data, in which visual range is reported for only a limited set of markers. With airport data, we can summarize the percentage of time that visual range is less than or greater than each marker distance, but we would have to interpolate to estimate the percentage of time that the visual range is less than or greater than a distance for which there is no marker. Early papers on visibility have all used this form of analysis (for example, Holzworth and Maga, 1960; Green and Battan, 1967; Duckworth and Kinney, 1978 and 1983; Latimer et al., 1978; and Miller, 1972). An example of a visibility trend analysis that uses the flux method is shown in Figure 4-1. We see that there are markers at 4, 7, 10, 12, 20, and 25 miles. The left vertical axis shows the percentage of time that visual range is less than the marker distances, and the right vertical axis Indicates the percentage of time that visual range is greater than the marker distances. Thus we see that in the summer quarter of 1973, visibility was less than 7 miles about 33 percent of the time, while in 1975 it was less than 7 miles about 25 percent of the time. The plot shows the main advantage of the flux method: With limited visibility data, such as that reported at airports, the plot shows exactly how the distribution of values changes across time. Since the flux method uses summary statistics that are unitless, actual trends in visual range across time are not reported, as would be the case with trends calculated using means or medians. Thus the flux plots are slightly more difficult to interpret. Another difficulty in inter- pretation can arise if there are different trends at different markers. This might erroneously appear to be the case if reporting practices change. For example, in Figure 4-1, we see that visibility in the winter quarter is not reported as greater than 25 miles until 1975. This may 87110 >» ------- Winter Quarter a> 01 c S. 5 (/> o IE 90 80 70 60 50 30 20 10 •o 20 30 40 50 50 70 SO 30 -972 1974 1976 1978 19SO Year c o a: in > 0> in u) 0> _l £! i O X 90 80 70 60 50 4-0 30 1962 i984 1386 Summer Quarter :o 20 30 40 50 SO 70 SO 30 ft Q i972 1974 1976 1978 1380 Year 1952 1984- 1386 Noontime Visibility at Baltimore, 1972-1986 FIGURE 4-1. Example of visibility trends characterized using flux method. 87110 32 ------- indicate a real change in the percentage of days with "good" visibility or it may indicate a change in the reporting practice. The major drawback of the flux method is that site-to-site comparisons are very difficult, if not impossible. If the objective is to compare the percentage of time visibility is above or below a reference value at each site, then plotting the frequency of good or bad visibility conditions for all sites would require interpolation and extrapolation of percentiles at some sites. Thus, to achieve spatial comparability one would be forced to interpolate and extrapolate from the cumulative frequency distribution, the avoidance of which is one of the principal advantages of the flux method. However, many researchers believe that it would be better to pre- sent visibility trends in a site-specific manner in any case, because of other problems associated with pooling data across stations. THE PERCENTILE APPROACH The most commonly used statistical indicator of visibility frequency dis- tributions is a percentile of the visibility frequency distribution. The 50th percentile, or median visibility, is an indicator of typical visibil- ity conditions: Half of the time visibility is better than the median value and half of the time it is worse. The 10th percentile is that vis- ibility which is exceeded 90 percent of the time. Similarly the 90th vis- ibility percentile is visibility that is exceeded 10 percent of the time during the period of record. Percentiles are currently used by most visi- bility trend analysts (e.g., Malm and Molenar, 1984; Sloane, 1981, 1982a,b, 1983, 1984; Trijonis and Yuan, 1978, 1979; Trijonis and Shapland, 1979; Trijonis, 1980). Percentiles are straightforward to understand and interpret since the units of the percentile are the same as the quantitative parameter used to describe visibility (i.e., the visual range or extinction coefficient); trend lines can therefore be drawn with y-axes in the units of interest. The extremes of the visibility frequency distribution (both good and bad visibility) can be characterized by different percentiles. 10th and 90th percentiles are most commonly used to characterize the tails of the dis- tribution; however, 25th and 75th percentiles could be used in their place, or in addition. Several percentiles can be effectively plotted on the same graph using the boxplot technique currently used in the EPA trends reports for summarizing national distributions of annual air qual- ity summary statistics. An example of visibility percentiles displayed in boxplots is provided in Figure 4-2. The median is indicated by the line in the center of each box; the top and bottom of each box indicates the 75th and 25th percentiles, and the black bars outside the boxes mark the 10th and 90th percentiles. 87110 33 ------- Winter Quarter 20 - 10 - 30 20 - u> - ; i - - - - ~ ^ 1 T 1972 1 T i H 1974 ,'• 1 ' 1 ' 1976 N 1 ' 1 I 1978 1 ^_ T 1 1 ^**. T 1980 -N 1 _• Y "" . I 1932 PO— T f' , I i 1984 ,_ 1 i" I - 0.9 L2 C7 X 1.5 5 — _> 2.4 ^ 4.9 986 Year Summer Quarter - - - I 1 i 1 A 1 1 I 1 1 1 1 I I - 0.9 1.2 o- 3_ X 1.5 5 10 - 2.4 4.9 1972 1974 1976 1978 1980 1982 1984 1986 Year Boxplot Comparisons of Visibility for Baltimore, 1S72 —1936 FIGURE 4-2. Example of visibility trends characterized using percentiles displayed with box plots. The percent!les displayed are the 10th, 25th, 50th (dashed line connects median across time), 75th, and 90th. 87110 34 ------- The main disadvantage of the percent!le approach when applied to U.S. Wea- ther Service airport visibility observations is that the calculation of specific percentiles, as typically carried out, requires interpolation of the empirical frequency distribution. This in turn requires assumptions about the form of the underlying frequency distribution. For example, assume that the markers used to report visual range at a given airport are at 5, 10, 30, 40, and 50 miles, and that the frequency of visual range at these markers is as follows: Cumulative Marker Frequency of Frequency of (miles) Exceedance (%) Exceedance (%) 50 5 5 40 15 . 20 30 10 30 10 30 60 5 40 100 What would the 50th percentile be? Figure 4-3 illustrates two possible answers. The actual cumulative percentages from the tabulation just given are shown as points on the plot. The 'step1 function shown by the dashed lines is statistically correct in the sense that it admits there is no data betwen points and assumes no more than the data record. The median calculated with this representation of the distribution is 30 miles. The common approach is not this step function, but rather an interpolation from the solid line connecting the reported values; as shown in the fig- ure, the 50th percentile or median calculated in this way is about 17 miles. Such an interpolation assumes a very simple model for the reported data—that frequencies of occurrences are uniform between markers. For example, 30 percent of the values are reported as greater than or equal to 10 miles but less than 30 miles. The interpolation scheme assumes that if we had markers at every mile, we would have exactly as many occurrences of 10 miles as we would of 11 miles or 12 miles or 29 miles. Clearly this is a naive assumption; several researchers have shown that visibility distri- butions, like most air quality distributions, are skewed (e.g., Husar et al., 1979). A second problem with the percentile method involves the estimation of extreme percentiles of the visibility frequency distribution. Suppose that there were no marker at 50 miles in our example tabulation, and sup- pose we were interested in the 90th percentile, as an indicator of good 87110 35 ------- ioo - c> e •> o> IB 40 - 20 - 2O 3O Marker Distance (miles) 4O 50 FIGURE 4-3. Example of calculation of percentiles in a visibility frequency distribution—interpolation (solid line) versus empirical step function (dashed line). 36 ------- visibility. We know only that visibility is better than 40 miles 20 per- cent of the time. To estimate the 90th percentile, we have to assume some mileage for the best possible visibility to be able to interpolate, or in this case really extrapolate, what the 90th percentile would be. Obvi- ously, the choice of "best" determines what the percentile will be. If the best visibility is assumed to be 60 miles, then the 90th percentile will be calculated as 50 miles; if the best visibility is assumed to be 70 miles, then the 90th percentile is 55 miles. At sites where visibility markers are few and far between, such interpolation and extrapolation can introduce significant uncertainty. A discussion of sources of uncertainty in these circumstances can be found in Sloane (1984, Appendix B). Some researchers have concluded that lognormal distributions fit visi- bility data in many cases. Husar et al. (1979) maintain there is con- siderable evidence that visibility and light extinction, like atmospheric concentrations of fine particles, are lognormally distributed. They inspected several locations and seasons for which entire visibility cumu- lative frequency distributions could be determined and found no evidence of gross inconsistency between the actual distributions and the lognormal distribution in the 10th to 90th percentile range. They developed a semi- automated extrapolation procedure whereby the cumulative frequency distri- bution function is fitted with a straight line (on lognormal probability paper), using a semicircular weighting function centered at 50 percent, with cutoffs at 10 and 90 percent. The fitted distributions were used to estimate the mean and percentiles. Other common statistical distributions, such as the Weibull, gamma, or inverted gamma, could also be fit to visibility data. The small number of routinely reported visual ranges at airports would not likely allow one to conclude which of these is a more accurate representation of the actual visibility distribution. More than likely, the data could appear to fit any number of distributions. Fitting a distribution to the data can cause potential problems in evalua- ting variations in visibility. In our example, different distributions cause the resulting median to vary by 12 miles (from 17 miles for the interpolated percentile to 30 miles using the step function). In effect, 13 miles of 'noise' are introduced into the analysis. This "noise" makes it difficult to determine whether a given annual or seasonal change (or lack of change) was a true change or just an artifact of the method of determining percentiles. 87110 U 37 ------- RIDIT ANALYSIS Another approach to obtaining a concise representation of the entire visi- bility frequency distribution for the purpose of trend analysis is through ridits. A ridit is the probability that a visibility observation in a given period was better or less than a visibility observation from a ref- erence distribution. The given period could be any year, while the refer- ence could be all years included in the trend analysis. Ridit analysis has been used by Craig and Faulkenberry (1979), Sloane (1982a), and Vinzani and Lamb (1985) for visibility trend analysis. The technique has the advantage of being able to use the specific visibility target dis- tances directly. It also uses the entire frequency distribution. A sin- gle number, the mean ridit, is obtained for each year. However, the approach has several disadvantages. The concept is more difficult to understand; a unitless probability that a given frequency distribution is better than a reference is less instructive than a percentile or mean visibility or extinction. Also, though the entire frequency distribution is used, ridit analysis is less sensitive to changes at the tails of the distribution; therefore the trends in 90th percentile (poor) visibility conditions may be underrepresented. Sloane (1982a) found that ridit anal- ysis is consistent with trend analysis based on percentiles (carefully applied) and therefore ridit analysis may offer no clear advantage over the percentile approach. Sloane1s more recent (1984) visibility trends work has used percentile visibility rather than ridit analysis. 87110 >» 38 ------- ALTERNATIVE METHODOLOGIES FOR INCORPORATING METEOROLOGICAL VARIABILITY IN VISIBILITY TREND ANALYSIS For the purposes of the national air quality trends report, it is desir- able to distinguish visibility trends associated with anthropogenic (human-caused) factors from those associated primarily with natural events. For example, low visibility may be due to natural events such as rain, snow, fog, windblown dust, or even volcanic eruptions. Other nat- ural factors, such as humidity, act in conjunction with existing pollutant concentrations to improve or degrade visibility. This section reviews methodologies used for incorporating meteorological variability in past visibility trend studies. SELECTION OF HOURS Airport visibility data for every third hour are archived in the National Climatic Data Center's data base. Thus, all daytime values (e.g., LST values for 0700, 1000, 1300, and 1600) can be used to compile visibility frequency distributions. Given 4 observations per day for 365 days, a total of 1460 data points per year would be available. Sloane (1982) has documented that certain hours of the day have generally bad visibility for reasons unrelated to pollutant levels. For example, many areas experience early morning fog. Furthermore, early in the morn- ing, high local particulate concentrations caused by trapping within the ground-level nocturnal inversion may cause visibility impairment. Sloane (1982a) noted that early morning (0700) observations did not follow the same pattern as did other daylight observations made at other times of day. Other times of day are subject to such influences as sun angle, which can result in visibility degradation because of increased backward or forward light scattering. Latimer et al. (1978), Trijonis and Yuan (1978, 1979), Sloane (1982a), Holzworth and Maga (1960), and Miller (1972) have all used all daytime visibilities in their trend analyses. Such an approach need not make assumptions regarding the contribution to visibil- ity improvement or impairment made by particular hours. Other researchers have used a single hourly visibility value, representing either midday or afternoon, rather than all daytime, values (Husar et al., 87110 5 3g ------- 1979, 1981, 1984; Trijonis and Shapland, 1979; Trijonis, 1980; Vinzani and Lamb, 1985). Because daily variations in visibility are most pronounced in the early morning or late afternoon, the use of midday or early after- noon observations may provide a data base that is less influenced by nat- ural meteorological phenomena. Vinzani and Lamb (1985) used afternoon (1500) data because temperatures are at their highest at that time and thus relative humidities are at their lowest and atmospheric mixing is at its highest. Some studies have used only those daylight hours during which natural visibility impairment would be unlikely to occur (Sloane, 1980, 1982; Malm, 1981). USE OF RUNNING AVERAGES TO SMOOTH DATA Because of year-to-year variation in meteorological conditions, many researchers have averaged visibility statistics from several years (most typically three) to obtain a running average (overlapping mean, of several years' statistics) in an attempt to smooth out interannual variations (Trijonis and Yuan, 1978, 1979; Husar et al., 1979; Sloane, 1982a, 1984; Vinzani and Lamb, 1985). One disadvantage of this approach is that it introduces another layer of averaging that moves farther from the actual data and the very real interannual variation in visibility conditions caused by meteorological differences. Also, running averages of several years' data are not currently computed for other kinds of air quality data reported in the EPA annual trends report (although some short-term stan- dards are based on expected exceedances, normally calculated as running averages of three years). ELIMINATION OF HOURS SUBJECT TO NATURAL VISIBILITY IMPAIRMENT An analysis of visibility trends related to anthropogenic factors may need to exclude from the data hours subject to natural visibility impairment. For example, precipitation is a natural phenomenon that limits visibility and is not representative of the pollutant concentration in an air mass. A more complex example is fog, which may form from natural or anthropo- genic aerosols and is sometimes difficult to distinguish from haze. Another meteorological factor, relative humidity, can influence the amount of impairment caused by a given particulate concentration. Other natural phenomena such as windblown dust can also result in visibility impairment unrelated to anthropogenic emissions. Almost all meteorological adjust- ment procedures entail removing hours subject to certain meteorological conditions. 87110 S 40 ------- Precipitation All trends studies we are aware of that have used meteorological adjust- ments have eliminated hours with precipitation from their analysis. Humidity. Fog, and Visibility In general, higher humidities result in greater visibility impairment. At higher relative humidities, more liquid water is associated with such hygroscopic and/or deliquescent aerosols as sulfates and nitrates and some organics, thereby increasing the light scattering mass. Sloane (1983, 1984) has shown that relative humidity within a region can vary signifi- cantly from one year to another, and that part of the observed downward trend in visibility in the 1950s can be attributed to climatic changes in humidity. Because humidity-related visibility impairment may be associa- ted with climatic changes rather than with increased pollution, some studies have deleted humidities greater than certain values such as 90 percent (Sloane, 1980, 1982a,b; Vinzani and Lamb, 1985; Holzworth and Maga, 1960; Miller, 1972; Craig and Faulkenberry, 1979) or 70 percent . (Duckworth and Kinney, 1978, 1983). A key factor in using a 90 percent cutoff point for humidity observations is that natural fog is unlikely to occur at relative humidities lower than 90 percent. Thus the elimination of relative humidities greater than 90 percent will eliminate most natural fog and preserve observations of haze that may have been mistakenly iden- tified as fog (Holzworth and Maga, 1960; Sloane 1982a). Reiss and Eversole (1978) presented a technique whereby visibility statis- tics can be adjusted to remove relative humidity effects on the basis of a semi-empirical relationship between the scattering coefficient and humid- ity. Another approach to removing relative humidity effects is to strat- ify the visibility data base by relative humidity and to compare only interannual visibilities within a given relative humidity class. TREATMENT OF METEOROLOGICAL VARIABILITY BY OTHER RESEARCHERS As Table 3-1 illustrates, existing studies have used many different approaches in analyzing and presenting visibility trends. The majority eliminate no data, or eliminate data for days with precipitation, fog, and/or humidity over 90 percent. There are some indications that trends are the same whether or not meteorological screening is performed. For example, Sloane (1980, 1982a) performed a trend analysis for Charlotte, North Carolina under seven different data-screening scenarios and found no statistically significant differences once early morning hours were elim- inated from the data. Vinzani and Lamb (1985) found that the inclusion of 87110 5 ., ------- observations of precipitation and high humidities in Evansville, Indiana did not affect visibility trends. A more complex adjustment for year-to-year meteorological variability used by Holzworth and Maga (1960) and others (Zeldin and Meisel, 1978; Chock et. alt 1982; Sloane, 1984) involves placing visual range values in dif- ferent meteorological categories based on standard wind speed, direction, and stability class. Pollutant levels can then be adjusted to reflect what they would be if, in each year, the frequency of occurrence of each meteorological class were typical. Alternatively, meteorological condi- tions can be classified by their impact on visibility and trends in visual air quality can be analyzed for each meteorological class separately or by normalizing the frequency of occurrence for each class. 87110 5 42 ------- PROTOTYPE ANALYSIS To provide a further basis for the conclusions and recommendations pre- sented in Section 7, we performed a detailed analysis of seven years of visibility observations from the Lambert International Airport in the St. Louis metropolitan area. This prototype analysis examined the (1) Choice of hours to be included in the trend data set, (2) Choice of visibility categories for use in flux analysis, and (3) Effect of techniques for incorporating meteorological variabil- ity in trend analysis. CHOICE OF HOURS The NWS reports visibility and meteorological data at least once every three hours. Nighttime visibility is defined and measured differently from daytime visibility and has little relationship to the public's per- ception of prevailing visual air quality. Therefore, we considered only daytime observations as candidates for analysis. Daytime observations are recorded at local times corresponding to 9:00 a.m., 12:00 noon, and 3:00 p.m. To reduce processing costs, to simplify the display of visibility pat- terns, and to avoid the inclusion of daytime hours that may have charac- teristically good or poor visibility for reasons unrelated to pollutant concentrations, trend analyses can be conducted for just one of the three available hours rather than for each of the hours or for their combined data. Figure 6-1 shows a plot of 10th percentile (poor) visibilities against time for each of the three daytime hours for which data are available. The patterns for each of the hours track each other well. Similar plots were examined for median and 90th percentile (good) visibility; again the trend lines were very similar. These analyses indicate that for this particular station, little information is lost by using data for only one hour per day. Analyses reported in the literature indicate similar 87110 6 43 ------- HK 9A.M. -M- 12 Noon -»- 3P.M. 1976 1977 1978 1979 1980 Year 1981 1982 1983 FIGURE 6-1. Trends in 10th percent!le (poor) visibility by recording hour at the St. Louis airport. 44 87110 ------- results (Sloane, 1982; Vinzani and Lamb, 1985). However, a more complex technique employed by (Sloane, 1984) suggested that meteorologically adjusted trends were more optimistic than analyses which did not consider meteorological factors and an error analysis. CHOICE OF VISIBILITY CATEGORIES FOR FLUX ANALYSIS Flux analysis examines trends in the frequency of occurrence for various categories of visibility. Correct definition of these categories requires an analysis of the annual frequencies of reported visibility at each sta- tion for reasons explained in the following paragraphs. As discussed in Section 3, visual range observations at a given station may change over time due to personnel or procedural changes, or because of changes in marker locations or reporting practices. These changes can result in inconsistencies in the measurements recorded, even if the true visibility remains constant. Such factors may lead to distortions in the visibility distribution for some time periods relative to others. If these distortions are serious, recorded measurements may obscure real trends in visibility or may give evidence of trends in visibility that are artifacts of reporting changes. Thus visibility categories used in repre- sentations of visibility data must be selected carefully to reduce the influence of procedural changes on the measurement of trends. For the station used in the exploratory analysis, the only official mar- kers used were at 3 and 4 miles, though over the 7-year period of record examined, observations were recorded at 18 different distances. Table 6-1 shows the frequency of observations by year at each of these 18 distan- ces. Distances are listed horizontally; years are listed vertically. The first number in a given box is the number of observations for that distance in that year. The next number is the percent of all observations reported by the first number. The third number is the percentage of times that distance was recorded in that year, relative to the other years. The fourth number is the percentage of a years observations at the distance. Prior to 1979, there were only a few isolated reports of visibilities greater than 10 miles, but after 1979 more than 10 percent of the observations in any year were recorded at distances of 11 miles or greater. Also noteworthy is the dramatic decrease in 1979 in the fre- quency of observations recorded at 7 miles and an increase in the number of observations reported at 8 miles. These changes imply a change in reporting practices rather than an actual change in visibility. Visibility categories should be chosen to reduce the sensitivity of the analysis to such procedural changes. In the example just given, the observations recorded at 8 miles could be combined with those recorded at 87110 6 45 ------- TABLE 6-1. Frequency of visibility by year at 18 distances for the St. Louis airport. Vis Year FREQUENCY PERCENT ROW PCT COL PCT 0.312 1.375 1.5 2 2.5 3 j 76 h -.--__H 1 0.01 100.00 0.10 0 0.00 -0.00 0.00 0 0.00 0.00 0.00 1 0.01 10.00 0.10 5 0.07 18.52 0.52 4 0.06 5.13 0.42 77 0 0.00 0.00 0.00 0 0.00 0.00 0.00 2 0.03 50.00 0.23 3 0.04 30.00 0.34 6 0.08 22.22 0.68 17 0.24 21.79 1.94 78 0 0.00 0.00 0.00 0 0.00 0.00 0.00 1 0.01 25.00 0.12 1 0.01 10.00 0.12 6 0.08 22.22 0.69 23 0.32 29.49 2.65 79 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 2 0.03 20.00 0.23 6 0.08 22.22 0.70 12 0.17 15.38 1.39 80 0 0.00 0.00 •o.oo 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 1 0.01 3.70 0.11 2 0.03 2.56 0.22 81| 82 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 5 0.07 6.41 0.54 0 0.00 0.00 0.00 1 0.01 100.00 0.11 1 0.01 25.00 0.11 1 0.01 10.00 0.11 1 0.01 3.70 0.11 13 0.18 16.67 1.48 83 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 2 0.03 20.00 0.23 2 0.03 7.41 0.23 2 0.03 2.56 0.23 TOTAL 953 877 868 862 904 924 879 861 13.37 12.30 12.18 12.09 12.68 12.96 12.33 12.08 TOTAL • 1 0.01 • 1 0.01 • 4 0.06 • 10 0.14 • 27 0.38 • 78 1.09 7128 100.00 Continued 87110 46 ------- TABLE 6-1. Continued. Vis FREQUENCY PERCENT ROW PCT COL PCT 4 5 - - -H 6 7 8 9 TOTAL 76 15 0.21 11.11 1.57 >.__ _.__H 29 0.41 8.71 3.04 58 9.57 6.09 87 . 1.22 5.92 9.13 120 1.68 21.16 12.59 1 0.01 4.17 0.10 953 13.37 77 22 0.31 16.30 2.51 52 0.73 15.62 5.93 86 14.19 9.81 122 1.71 8.30 13.91 119 1.67 20.99 13.57 0 0.00 0.00 0.00 877 12.30 78 23 0.32 17.04 2.65 66 0.93 19.82 7.60 68 11.22 7.83 104 1.46 7.07 11.98 154 2.16 27.16 17.74 0 0.00 0.00 0.00 868 12.18 79 26 0.36 19.26 3.02 53 0.74 • 15.92 6.15 123 20.30 14.27 325 4.56 22.11 37.70 53 0.74 9.35 6.15 5 0.07 20.83 0.58 862 12.09 Year 80 14 0.20 10.37 1.55 39 0.55 11.71 4.31 62 10.23 6.86 154 2.16 10.48 17.04 57 0.80 10.05 6.31 11 0.15 45.83 1.22 904 12.68 81 8 0.11 5.93 0.87 20 0.28 6.01 2.16 64 10.56 6.93 199 2.79 13.54 21.54 36 0.51 6.35 3.90 7 0.10 29.17 0.76 924 12.96 82 16 0.22 11.85 1.82 40 0.56 12.01 4.55 56 9.24 6.37 204 2.86 13.88 23.21 15 0.21 2.65 1.71 0 0.00 0.00 0.00 879 12.33 83 11 0.15 8.15 1.28 34 0.48 10.21 3.95 89 14.69 10.34 275 3.86 18.71 31.94 13 0.18 2.29 1.51 0 0.00 0.00 0.00 661 12.08 TOTAL 135 1.89 333 4.67 606 1470 20.62 567 7.95 24 0.34 7128 100.00 Continued 47 87110 ------- TABLE 6-1. Concluded. Vis HREQUENCY PERCENT ROW PCT COL PCT 10 12 13 15 20 - - .._„-.- H 25 TOTAL 76 333 4.67 14.61 34.94 299 4.19 42.78 31.37 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 953 13.37 77 270 3.79 11.84 30.79- 172 2.41 24.61 19.61 0 0.00 . 0.00 0.00 6 0.08 0.79 0.68 0 0.00 0.00 0.00 - - - - - - H 0 0.00 0.00 0.00 877 12.30 78 283 3.97 12.41 32.60 139 1.95 19.89 16.01 0 0.00 0.00 0.00 0 0.00 0.00 0.00 0 0.00 0.00 0.00 h - - _._-H 0 0.00 0.00 0.00 868 12.18 Y 79 211 2.96 9.25 24.48 18 0.25 2.58 2.09 4 0.06 7.41 0.46 23 0.32 3.04 2.67 1 0.01 1.25 0.12 K - - . ..._H 0 0.00 0.00 0.00 862 12.09 ear 80 266 3.73 11.67 29.42 32 0.45 4.58 3.54 30 0.42 55.56 3.32 211 2.96 27.87 23.34 23 0.32 28.75 2.54 h---- _.-.H 2 0.03 100.00 0.22 904 12.68 81 294 4.12 12.89 31.82 13 0.18 1.86 1.41 19 0.27 35.19 2.06 209 2.93 27.61 22.62 50 0.70 62.50 5.41 0 0.00 0.00 0.00 924 12.96 82 287 4.03 12.59 32.65 25 0.35 3.58 2.84 0 0.00 0.00 0.00 213 2.99 28.14 24.23 6 0.08 7.50 0.68 0 0.00 0.00 0.00 879 12.33 83 336 4.71 14.74 39.02 1 0.01 0.14 0.12 1 0.01 1.85 0.12 95 1.33 12.55 11.03 0 0.00 0.00 0.00 0 0.00 0.00 0.00 861 12.08 TOTAL 2280 31.99 699 9.81 54 0.76 757 10.62 80 1.12 2 0.03 t- 7128 100.00 87110 48 ------- 7 miles since a visibility recorded at 7 miles means that visibility was 7 miles or greater. For the test site examined, 10 miles is apparently the lowest upper limit used in reporting visibilities over the period, so "10 miles or greater" would be the category used for any observation recorded at any distance greater than 10 miles. On the basis of the frequency table and informa- tion provided by the observation station, visibility cutoffs of 3, 4, 7, and 10 miles (that is, less than 3 miles; at least 3 miles but less than 4 miles; at least 4 miles but less than 7 miles at least 7 miles but less than 10 miles; and 10 miles or further) were selected for the sample anal- yses for this station. INCORPORATION OF METEOROLOGICAL VARIABILITY As discussed in Section 5, many visibility trend studies utilize some form of screening on the basis of meteorological measurements to present trends that reflect the impact of human activities on visibility. To gain more insight regarding the best way to accomplish this screening, our prototype analysis examined (1) The distribution of visibilities under various screening scenarios, (2) The proportion of data that would be omitted under various screening scenarios, (3) The distribution of visibility in different humidity categories, and (4) 10th, 50th and 90th percentile visibilities for five different humidity categories. We analyzed the distribution of visibility under five different screening scenarios: one for all daytime observations; one for daytime hours with no precipitation; one for only those daytime hours with precipitation; one for daytime hours with no fog or other nonanthropogenic aerosols; and one for only those hours with nonanthropogenic aerosols. A dramatic decrease in the frequency of poor visibilities was noted upon exclusion of hours with precipitation, which indicates a strong association between precipi- tation and poor visibility. There is a greater frequency of poorer visi- bilities when hours with natural aerosols are included than when they are excluded. Table 6-2 shows the proportion of data by season and year that would be omitted under the screening procedures recommended in Section 7. The total percentage of data thus excluded ranges from 10.9 percent 87110 6 4g ------- TABLE 6-2. Percentage of noontime observations excluded from analyses for prototype station due to adverse meteorological conditions. Year 1976 1977 1978 1979 1980 1981 1982 1983 Winter 12.2 24.4 33.3 41.1 33.3 10.0 31.1 40.0 Spring 12.0 12.0 25.2 25.0 14.1 15.2 8.7 25.0 Season Summer 5.4 8.7 6.5 4.3 7.6 8.7 8.7 3.2 Fall 14.3 23.0 16.5 12.0 9.9 14.2 19.8 3.3 Annual 10.9 17.0 20.3 20.5 16.1 12.1 17.0 20.8 871102 50 ------- for 1976 to 20.8 percent for 1983. The percentage omitted by season is more dramatic. In summer and fall the percentage of omitted data does not differ greatly from the annual amount. In winter and spring, more data are omitted, almost twice as much in 1979 as the highest amount in 1983. As discussed earlier, poor visibility is often associated with high humid- ity. In California, the visibility standard incorporates humidity effects. A violation of the California standard for visibility occurs when the concentration of visibility-reducing particles is sufficient to reduce visibility to less than 10 miles when humidity is less than 70 per- cent. At the St. Louis station, however, eliminating any observations for which humidity was greater than 70 percent would remove over 30 percent of the data from the record. This suggests that the California approach is not appropriate for all parts of the country. The elimination of hours with higher humidities may also eliminate hours with higher particulate concentrations; high humidities coupled with high temperatures may result in increased emissions of visibility-reducing particles due to increased demand for electric power for air conditioners. Our analysis showed a strong association between adverse visibilities and high humidities. Visibility distributions do not appear to be associated with humidity at humidities lower than 60 percent, but are highly correlated with humidi- ties greater than 60 percent. A key issue in eliminating hours with high relative humidities is the question of whether such screening will change the trend in visibili- ties. Figure 6-2 shows 10th percentile visibilities as a function of time for each of five levels of humidity screening: no screening, removal of hours with humidities greater than 95 percent, 90 percent, and 85 per- cent. Although there are some differences in the trend lines between the different levels of screening, they track each other remarkably well. Similar graphs were examined for the median and 90th percentile visibili- ties; the effects of humidity screening on the middle and high ranges of the visibility distribution are negligible. It would be incorrect to generalize these results to other stations or for longer periods of time, but the results for the prototype station do not provide evidence that humidity screening would have a dramatic effect on the visibility patterns detected. As described earlier, however, an analysis by Sloane (1983, 1984) has shown that changes in humidity may have contributed signifi- cantly to the observed downward trend in visibility in the late 1950s. 871106 51 ------- •+• No Humidity Scr •*- RH<95% -M- RH<90% -•- RH<85% 1976 1977 1978 1979 1980 Year 1981 1982 1983 FIGURE 6-2. Trends in 10th percentile (poor) visibility using different meteorological screening procedures. 87110 52 ------- CONCLUSIONS AND RECOMMENDATIONS This section summarizes our conclusions and recommendations for developing visibility trends data to be included in the National Emissions and Air Quality Trends Report. Recommendations cover the visibility data base, visibility indicator, geographical areas for study, time period, and method of accounting for natural visibility impairment. Recommendations are based on the literature review and telephone survey of visibility experts conducted as part of this study, and on our own analysis of data from St. Louis, Missouri. Table 7-1 summarizes the advantages and dis- advantages of each recommendation. RECOMMENDED DATA BASE Our research indicates that the only data base currently appropriate for use in visibility trends analyses is a compilation of NWS airport visibility observations. Since no other data base approaches its geographical extent and long period of record, we recommend that these data be used in characterizing visibility trends for the National Trends Report. A careful selection of stations, visibility indicators, and presentation format can reduce the limitations caused by variations in marker availability and observer practices at different airports. RECOMMENDED VISIBILITY INDICATOR We recommend that visual range be used for trend analysis and reporting. Although there are advantages and disadvantages associated with the use of any indicator, visual range is most closely related to people's everyday experience and would make the trends report useful to the largest number of people (Bachmann, 1987). Many researchers prefer bext because it is directly related to aerosol concentrations and can easily be converted to visual range through simple relationships such as the Koschmeider equa- tion. However, the assumptions implied in the Koschmeider equation are often not valid and, as discussed earlier, some research has suggested that different relationships would apply to different situations. If we find that the percentile analysis is the most useful approach for describ- ing visibility, we will present graphs showing visual range on the left y-axis and bext on the right. 87110 7 53 ------- TABLE 7-1. Advantages and disadvantages of components recommended for visibility trends analysis. Trends Analysis Components Recommendation Advantages Disadvantages Data base Visibility Indicator Statistical parameter NWS airport data Visual range 10th. 25th, 50th. 75th. 90th percent lie visibility Percent frequency less than various distances Long period of record. Large spatial No data for Class I areas. Visual range 1s coverage. Use for trends Is well the only visibility Indicator reported; some established differences among airports Closest to what people see. to what airport data report Closest Consistent with current Trends Report. Makes Interslte compari- sons easier Reports only what the data show; does not require Interpolation of data Not a good Indicator for visual features such as contrast and texture. Requires assumptions to relate it to atmospheric pollutant concentrations Hust Interpolate most percentlles from reported data Visibility categories may change over time due to changes In reporting practices or marker availability, making 1t diffi- cult to compare data between years. Inter-site comparisons are difficult en Temporal features Noontime observations Annual, winter, summer 15-year trend time period Uses the hour least likely to have natural impairment Analyzes seasons with most striking trends while providing annual sum- maries Longer than the 10-year suggested minimum; uses data after 1970, when reporting practices changed Does not use all available daylight observations Other seasons may also have Important trends Not consistent with trend time period currently used 1n Trends Report. Does not use all available data Meteorological screening Geograhlcal Area Delete data for hours with fog when relative humidity > 85%. hours with wind- blown dust, and hours with precipitation 10 urban sites Screens out data that may be more representative of natural Impair- ment than anthropogenic Impairment. Leaves in data for high humidities that also may have anthropogenic impairment Includes 10 of the 14 metropolitan areas covered In Trends Report. Sites in all but one EPA region Does not use all available data May not be adequate to characterize entire United States; 1n particular, not adequate to characterize rural visibility 87110 2 ------- CALCULATION AND DISPLAY OF TREND STATISTICS In Section 4, we discussed several methods of displaying trends in visi- bility data. Rather than recommend one specific approach at this time we recommend applying two of the approaches discussed. We will use the flux method to show trends in reported visibility data, and we will also calcu- late percentiles of the visibility distribution at each site and portray these in box plots. The percentiles of interest (10, 25, 50, 75, and 90) will be calculated by linear interpolation. For those cases in which linear interpolation is not possible at the extremes of the distribution, we will show a truncated box in the box plots rather than extrapolate a percentile. We will recommend which plots should be used in the national trends report after examination of both sets of plots. If both sets show the same gen- eral trends, then we will recommend box plots that are consistent with the data displays presented in the air quality sections of the Trends Report. GEOGRAPHICAL AREAS RECOMMENDED FOR STUDY In Section 1, four potential goals of visibility trends analysis were identified, each suggesting a specific geographical focus for analysis. A visibility trends study might examine trends in urban, rural, or Class I areas, or in some combination of these. The level of visibility in each of these three areas is important for a different reason. Most trends studies in the past have examined only rural and/or urban trends, largely because of the lack of a long-term data base for Class I areas. We recommend that the 1986 trends report contain visibility trends in urban areas and that future trends reports summarize trends for urban, Class I, and selected rural and other areas that may be of particular interest. Since large changes in emissions can quickly alter visibility in a region, future trends reports should examine the effects of such changes on rural or other areas. For example, S02 emissions have decreased by 40 percent in parts of California since 1979; Smelter shutdowns and controls in Arizona, Seattle, and Utah since 1970 have dramatically reduced S0£ emissions in these states; S02 emissions in the southeastern United States have increased sig- nificantly in the past 10 to 20 years. 871107 ------- Specifically, we recommend that visibility trends be presented for each of 10 metropolitan statistical areas chosen in conjunction with the EPA project officer. These areas are among the 14 currently used in the Trends Report. This will provide consistency with pollutant summaries and allow comparisons of visibility trends and trends in pollutant concentra- tions for these 10 areas. Until a longer period of record for Class I areas is available, it will be difficult to study visibility trends in these areas. When such a record is obtained, it will be necessary to formally resolve uncertainties aris- ing from the assumptions used to convert instrument measurements to visual range estimates. It should also be recognized that trends can only be developed for those Class I areas equipped with visibility monitors. The necessary level of detail for Class I areas in the Trends Report may be greater than is appropriate for urban or rural areas because of the spe- cial value of visibility in these areas. For example, visual range alone may be a convenient, usable measure of visibility trends in urban areas, but not in parks, where color and contrast may play more important roles. Thus it may be necessary to examine trends in Class I areas using more than one visibility indicator. RECOMMENDED TEMPORAL FEATURES OF TREND ANALYSIS The temporal components of a trend analysis include the overall trend time period (5, 10, 15, 50, years), the time period over which trends are sum- marized (monthly, seasonally, annually, etc.), and the times of day used in the analysis (daytime observations, noontime, etc). Meteorological variability is the fundamental consideration for choosing each of these components. Consideration of these components is necessary whether or not trends are to represent only anthropogenic factors because annual, sea- sonal, and diurnal meteorological variability can either imply anthropo- genic trends that do not exist or obscure trends that do. Overall Trend Time Period It is possible that during a period of 2 to 3 years unusually favorable or adverse meteorological conditions will occur that are not representative of typical conditions. As a result, it becomes necessary to choose an verall trend time period that brackets several such periods. For example, some visibility trends research based on airport data has analyzed trends over relatively short periods such as 7 years (Miller, 1972), but most studies have used at least 14 years (see Table 3-1 for examples). The results of our telephone survey indicate that 5 years is considered to be 87110 7 ------- present annual trends in addition to seasonal trends in order to examine trends over all data and to provide a condensed summary consistent with annual values reported in the Trends Report. We recommend presenting trends for summer and winter, along with annual summaries. We propose defining summer as July, August, and September, and winter as December, January, and February. Although the seasons thus defined do not correspond to the calendar quarters often used in the trends report, we believe that they best encompass those meteorological conditions most representative of each season (i.e., stability and wind speed). RECOMMENDATIONS FOR METEOROLOGICAL SCREENING PROCEDURES Techniques for eliminating data for hours subject to natural visibility impairment such as rain have been almost universally employed in studies examining anthropogenic visibility trends. Some of these techniques, such as elimination of hours with precipitation from the data, are not contro- versial; others, such as elimination of certain levels of humidity, are more debatable. As explained in Section 5, it is common to use a relative humidity level of about 90 percent as a cutoff point for the elimination of data due to humidity because of the way in which aerosols are formed and the difference between natural fog and anthropogenic haze. However, in California, the visibility standard is based on a 70 percent relative humidity level. We recommend deleting data for hours with fog when rela- tive humidity is greater than 85 percent. We recommend using 85 percent as a cutoff point on the basis of our preliminary investigation of the 10 urban sites selected for study. Our screening analyses showed a high degree of correlation between the reporting of fog as an impairment to visibility and an RH of 85 percent or greater. Data should also be deleted for hours during which there is precipitation or when windblown dust is recorded. This approach would remove from the analysis, to the extent possible, those hours characterized by visibility impairment due to natural phenomena. MINIMUM DATA REQUIREMENTS The EPA establishes minimum data requirements for sites to be included in an air quality trends analysis. These data requirements specify a minimum number of years of continuous data as well as the minimum amount of data in each year to be obtained. These requirements are essential because of the frequency of missing air quality data and the differing lengths of the monitoring records. For airport visibility data, however, we need not be as concerned about the effects of missing data. All airport data sets 87110 7 57 ------- under consideration have a data record of decades and data are rarely missing. However, we will omit observations when meteorological factors (fog, precipitation, or windblown dust) result in low visibility. There- fore, we need to establish minimum data requirements in case many observa- tions are excluded for meteorological reasons. We recommend that a month be considered representative if at least half of the days in the month are included in the analysis, and that a season be considered valid if all three months are representative. 87110 7 58 ------- REFERENCES Allard, 0., and I. Tombach. 1981. The effects of non-standard conditions on visibility measurement. Atmds. Environ., 15:1847. Bachmann, J. 1987. U.S. Environmental Protection Agency, OAQPS, ASB, personal communication. Bachmann, J., et al. 1985. "Developing Long-Term Strategies for Regional Haze: Findings and Recommendations of the Visibility Task Force." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina. Berg, N. 1986. "Status Report on Visibility Monitoring." U.S. Environ- mental Protection Agency. Blumenthal, D., and J. Trijonis. 1984. "Visibility Monitoring in the Southern California Desert for the Department of Defense: Research on Operations-Limiting Visual Extinction." Naval Weapons Center, China Lake, California (NWC TP 6566). DOC, DOD, DOT. 1979. Surface Observations. Federal Meteorological Hand- book No. 1, 2nd ed. U.S. Department of Commerce, U.S. Department of Defense, and U.S. Department of Transportation. Duckworth, S., and J.J.R. Kinney. 1978. "Visibility Trends in the Great Central Valley of California, 1958-1977." California Air Resources Board, Sacramento, California. Duckworth, S., and J.J.R. Kinney. 1980. "Visibility Trends in the Coastal Areas of California, 1958-1977." California Air Resources Board, Sacramento, California. Duckworth, S., and J.J.R. Kinney. 1981. "Visibility Trends in the Pris- tine Areas of California, 1958-1977." California Air Resources Board, Sacramento, California. 87110/3 59 ------- EPA. 1979. "Protecting Visibility: An EPA Report to Congress." U.S. Environmental Protection Agency, Research Triangle Park, North Caro- lina (EPA-450/5-79-008). EPA. 1987. "National Air Quality and Emissions Trends Report, 1985." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-450/4-87-001). EPRI. 1987. "Western Regional Air Quality Studies. Visibility and Air Quality Measurements: 1981-1982." Electric Power Research Insti- tute, Palo Alto, California (EPRI EA-4903). Espey, Huston & Associates, Inc. 1984. "Assessment of Urban and Rural Visibility Impairment in Texas." Prepared for Texas Air Control Board, Austin, Texas. Evans, E. G., T. R. Fitz-Simons, T. A. Lumpkin, and W. F. Barnard. 1987. "Establishment of an Eastern Visibility Fine Particle Net- work." 80th Annual Meeting of the Air Pollution Control Association, New York (21-26 June 1987). Green, C. R., and L. J. Battan. 1967. A study of visibility versus popu- lation growth in Arizona. J. Arizona Acad. Sci.. 4:226. Hartmann, W. K. 1972. Pollution: Patterns of visibility reduction in Tucson. J. Arizona Acad. Sci., 7:101. Holzworth, G. C., and J. A. Maga. 1960. A method for analyzing the trend in visibility. J. Air Pollut. Control Assoc., May 1960. Husar, R. B., J. M. Holloway, D. E. Patterson, and W. E. Wilson. 1981. Spatial and temporal pattern of eastern U.S. haziness: a summary. Atmos. Environ., 15:1919-1928. Husar, R. B., and D. E. Patterson, 1984. "Haze Climate of the United States." Prepared for the U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (CR 810351). Husar, R. B., and D. E. Patterson. 1986. "Haze Climate of the United States." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina. Husar, R. B., D. E. Patterson, J. M. Holloway, W. E. Wilson, and T. G. Ellestad. 1979. "Trends in Eastern U.S. Haziness Since 1948." Proc. Fourth Symposium on Turbulence, Diffusion and Air Pollution, pp. 249-256, American Meteorological Society, Boston, Massachu- setts. 87110/3 ------- Keith, R. W. 1970. "Downtown Los Angeles Noon Visibility Trends 1933- 1969." Air Pollution Control District, Los Angeles, California (Air Quality Report No. 65). Kelso, A. R. 1981. "Atmospheric Visibility Measurements at China Lake: A 5-year Nephelometry Summary." Naval Weapons Center, China Lake, California (NWC TP 6205). Kelso, R., R. Weiss, and A. Waggoner. 1984. "Measurements of Visual Air Quality at Five Sites in the Mojave Desert," Paper 84-59.5, presented at 77th Annual Meeting of the Air Pollution Control Association, San Francisco, California. Latimer, D. A., G. E. Anderson, H. Hogo, R. G. Ireson, R. E. Morris, and P. Saxena. 1984. "Visibility and Other Air Quality Benefits of Sul- fur Dioxide Emission Controls in the Eastern United States: Volume I." Systems Applications, Inc., San Rafael, California (SYSAPP- 83/248). Latimer, D. A., R. W. Bergstrom, S. R. Hayes, M. K. Liu, J. H. Seinfeld, G. Z. Whitten, M. A. Wojcik, and M. J. MilIyer. 1978. "The Development of Mathematical Models for the Prediction of Atmospheric Visibility Impairment." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-450/3-78-110a,b,c). Malm, W. C. 1979. Considerations in the measurement of visibility. J. Air Pollut. Control Assoc., 29:1042-1052. Malm, W. C. 1987a. National Park Service, personal communication. Malm, W. C., and R. C. Henry. 1987b. "Regulatory Perspective of Visi- bility Research Needs." 80th Annual Meeting of the Air Pollution Control Association, New York (21-26 June 1987). Malm, W., M. Pitchford, and A. Pitchford. 1982. Site specific factors influencing the visual range calculated from teleradiometer measure- ments. Atmos. Environ., 16(10):2323-2333. Malm, W. C., and E. G. Walther. 1980. "A Review of Instruments Measuring Visibility-Related Variables." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-600/4-80-016). Malm, W. C., and J. V. Molenar. 1984. Visibility measurements in national parks in the Western U.S. J. Air Pollut. Control Assoc., 34:899-904. 87110/3 61 ------- Mathai, C. V.t and I. H. Tombach. 1985. "Assessment of the Technical Basis Regarding Regional Haze and Visibility Impairment." Prepared for the Utility Air Regulatory Group by AeroVironment Inc., Monrovia, California. Mathai, C. V., and I. H. Tombach. 1987. A critical assessment of atmo- spheric visibility and aerosol measurements in the eastern United States. J. Air Pollut. Control Assoc.. 37:700-707. Middleton, P., and T. R. Stewart. 1985. On the use of human judgment and physical/chemical measurements in visual air quality management. J± Air Pollut. Control Assoc., 35(1):11. Middleton, P., T. R. Stewart, and D. Ely. 1984. Physical and chemical indicators of urban visual air quality judgments. Atmos. Environ., 18(4):861-870. Miller, M. E., N. L. Canfield, T. A. Ritter, and C. R. Weaver. 1972. Visibility changes in Ohio, Kentucky, and Tennessee from 1962 to 1969. Monthly Weather Review, 100:1. Molenar, J. 1987. Air Resource Specialists, personal communication. Mueller, P. K., J. G. Watson, J. Chow, K. Fung, and S. Heisler. 1982. "Eastern Regional Air Quality Measurements. Volume 1." Environ- mental Research & Technology, Inc., Concord, Massachusetts (EA-1914). John Muir Institute for Environmental Studies, Inc. 1983. "Western Regional Visibility Monitoring: Teleradiometer and Camera Net- work." U.S. Environmental Protection Agency, Las Vegas, Nevada (TS- AMD-8035b). Ozkaynak, H., A. D. Schatz, and G. D. Thurston. 1985. Relationships between aerosol extinction coefficients derived from airport visual range observations and alternative measures of airborne particle mass. J. Air Pollut. Control Assoc., 35:1176-1185. Patterson, D. E., J. M. Holloway, and R. B. Husar. 1980. "Historical Visibility over the Eastern U.S.: Daily and Quarterly Extinction Coefficient Contour Maps." U.S. Environmental Protection Agency (EPA-600/3-80-043a). Pitchford, M. 1987. U.S. Environmental Protection Agency, ESRL, Las Vegas, personal communication. 87110/3 62 ------- Pitchford, A., M. Pitchford, and W. Malm. 1981. Regional analysis of factors affecting visual air quality. Atmos. Environ., 15(10/11): 2043-2054. Reiss, N. M., and R. A. Eversole. 1978. Rectification of prevailing visibility statistics. Atmos. Environ., 12:945-950. Roberts, E. M., J. L. Gordon, D. L. Haase, R. E. Kary, and J. R. Weiss. 1974. "Visibility Measurements in the Painted Desert Through Photo- graphic Photometry." Paper presented at the 78th National Meeting of the American Institute of Chemical Engineers, Salt Lake City, Utah, August 1974. Sloane, C. S. 1982a. Visibility trends I: methods of analysis. Atmos. Environ., 16:41-51. Sloane, C. S. 1982b. Visibility trends II: mideastern U.S. 1948- 1978. Atmos. Environ.. 16:2309-2321. Sloane, C. S. 1983. Summertime visibility declines: meteorological influences. Atmos. Environ., 17:763-774. Sloane, C. S. 1984. Meteorologically adjusted air quality trends: visi- bility. Atmos. Environ., 18:1217-1229. Sloane, C. S., and W. White. 1986. Visibility: An evolving issue. Environ. Sci. Tech., 20(8):760. Sloane, C. S. 1987. General Motors Research Lab, personal communication. Sloane, C. S., and P. J. Groblicki. 1981. Denver's visibility history. Atmos. Environ., 15:2631-2638. Stonefield, D. H., and J. C. Metsa. 1987. "EPA's Visibility Program Up- date of the Phase I Rules." 80th Annual Meeting of the Air Pollution Control Association, New York (21-26 June 1987). Thompson, D. 1987. New York La Guardia Airport Observer, personal communication. Tombach, I., and D. Allard. 1983. "Comparison of Visibility Measurement Techniques: Eastern United States." Electric Power Research Insti- tute, Palo Alto, California (EPRI EA-3292). Tombach, I. H., D. W. Allard, R. L. Drake, and R. C. Lewis. 1982. "Western Regional Air Quality Studies: Visibility and Air Quality Measurements: 1981-1982." Electric Power Research Institute (EPRI EA-4903). 63 87110/3 ------- Trijonis, J., 1980. "Visibility in California." Technology Service Corp., Santa Monica, California (TSC-PD-B612-3). Trijonis, J. 1981. "Existing and Natural Background Levels of Visibility and Fine Particles in the Rural East." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-450/4-81-036). Trijonis, J. 1982. Visibility in California. J. Air Pollut. Control Assoc.. 32:165-169. Trijonis, J. 1984. "Visibility Research Review for the Texas Air Control Board." Santa Fe Research Corporation, Bloomington, Minnesota. Trijonis, J. 1987. Santa Fe Research Corporation, personal communica- tion. Trijonis, J., G. Cass, G. McRae, Y. Horie, W. Lim, N. Chang, and T. Cahill. 1982. "Analysis of Visibility/Aerosol Relationships and Visibility Modeling/Monitoring Alternatives for California." Cali- fornia Air Resources Board, Sacramento, California. Trijonis, J., and D. Shapland. 1979. "Existing Visibility Levels in the United States." U.S. Environmental Protection Agency, Research Tri- angle Park, North Carolina (EPA-450/5-79-010). Trijonis, J., and K. Yuan. 1978a. "Visibility in the Southwest: An Exploration of the Historical Data Base." U.S. Environmental Protec- tion Agency, Research Triangle Park, North Carolina (EPA-600/3-78- 039). Trijonis, J., and K. Yuan. 1978b. "Visibility in the Northeast: Long- term Visibility Trends and Visibility/Pollutant Relationships." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-600/3-78-075). Trijonis, J., and K. Yuan. 1979. Visibility in the southwest: an exploration of the historical data base. Atmos. Environ., 13:833- 843. Vinzani, P. G., and P. J. Lamb. 1985. Temporal and spatial visibility variation in the Illinois vicinity during 1949-80. J. Climate and Applied Meteorology, 24:435-449. White, W. 1986. On the theoretical and empirical basis for apportioning extinction by aerosols: A critical review. Atmos. Environ., 20:1659-1672. 87110/3 64 ------- White, W. 1987. Washington University, St. Louis, personal communica- tion. Wilson, W. 1987. U.S. Environmental Protection Agency, ASRL, personal communication. Zeldin, M. D.f and W. S. Meisel. 1978. "Use of Meteorological Data in Air Quality Trend Analysis." U.S. Environmental Protection Agency, Research Triangle Park, North Carolina (EPA-450/3-78-024). Zimmermann, K. A., and L. Wright. 1985. "Atmospheric Visibility Trends in Dallas-Fort Worth and El Paso, Texas." Paper 85-10.3, presented at 78th Annual Meeting of the Air Pollution Control Association, Detroit, Michigan, June 1985. 87110/3 ------- Appendix SELECTED INSTRUMENT-BASED VISIBILITY AND FINE PARTICLE MONITORING NETWORKS 87110 11 66 ------- Appendix SELECTED INSTRUMENT-BASED VISIBILITY AND FINE PARTICLE MONITORING NETWORKS INSTRUMENT-BASED VISIBILITY MONITORING NETWORK Several other instrument-based visibility monitoring efforts have been carried out in the recent past; however, each of these studies are limited in their spatial and temporal coverage. Some examples include the EPRI Eastern Regional Air Quality Study (ERAQS) and Western Regional Air Quality Study (WRAQS), the EPA Regional Air Pollutants Study (RAPS); the Subregional Cooperative Environmental Protection Agency, National Park Service, and Electric Utilities Study (SCENES); Research on Operations Limiting Visual Extinction (RESOLVE); the Pacific Northwest Regional Aero- sol Measurement Apportionment Study (PANORAMAS); and Visibility Impairment due to Sulfur Transport and Transformation in the Atmosphere (VISTTA). EPRI Eastern Regional Air Quality Study (ERAQS) For nine months in 1979 and 1980, visibility data were collected at two sites in Pennsylvania and Ohio using telephotometry, nephelometry, photography, and human observations (Tombach and Allard, 1983). EPRI Western Regional Air Quality Study (WRAQS) For 22 months in 1981 and 1982, visibility data were collected at 11 sites in the western United States using telephotometry, nephelometry, and human observations (Tombach et al., 1987). Aerosol sampling and analysis and meteorological measurements were also performed (EPRI, 1987). EPA RAPS The EPA operated a 20-site visibility and fine particulate network using a ground and aircraft-based measuring system in the St. Louis area for approximately three years from 1975 through 1977. Nephelometers were used to estimate the scattering coefficient. 87110 11 67 ------- SCENES Visibility and particulate measurements have been continuing since 1983 at 6 sites in the Southwest using teleradiometry, photography, nephelometry, and particulate samplers. This network has been operated jointly by the NPS, EPA, EPRI, Salt River Project, Southern California Edison, and Department of Defense. RESOLVE; Visibility measurements have been made since August 1983 at 7 sites in the California desert in an attempt to characterize contributing sources to regional haze at Department of Defense sites. Three sites were in source areas of Southern California; and four in receptor areas in the Mohave desert. The study used particulate samplers, nephelometers, and tele- radiometers. PANORAMAS A joint study by EPA Region X, and the states of Oregon, Washington, and Idaho consisting of 26 sites located in these states. Data was collected from July to the beginning of October in 1984 with nephelometers, fine particulate monitors, cameras, and a four telephotometers (Berg, 1986). Short-term intensives Visibility measurements have been made at a variety of locations for limited periods (much less than a year) as parts of various studies designed to evaluate visibility and/or aerosol conditions and relation- ships. Such short-term studies were carried out in Denver, Detroit, Hous- ton, Los Angeles, Shenandoah Valley, Research Triangle Park, Louisiana Gulf Coast, Atlantic Coast, and Great Smoky Mountains. Details are pro- vided in the review by Mathai and Tombach (1987). 87110 11 68 ------- FINE PARTICLE AND SULFATE MONITORING DATA National Air Surveillance Network (NASN) This network measured TSP and sulfur from 1956 to 1975 at 46 urban and 20 rural sites throughout the United States. Most of these data have been entered into the SAROAD system. EPA Inhalable Particulate Network (IP) This network measured particulate concentrations in three size ranges from 1978-1984: 0-2.5, 0-15, and 15-30 micrometers. Samples were collected at 163 sites in the United States. These data have known biases due to the use of glass fiber filters. EPRI SURE (Sulfate Regional Network) This network was operated by the EPRI at 54 stations throughout the north- eastern United States during 1977 and 1978. Only nine stations operated continuously; the remainder operated for only one month per season. Total suspended particulate concentrations and sulfate and nitrate fractions were measured. EPA Ohio River Valley (ORV) Aerosol Study: Three monitoring stations were operated from May 1980 to August 1982 in the Ohio River Valley. Both fine and coarse aerosols were collected. EPA Western Particle Characterization Study (WPCS); The EPA also collected fine and coarse particulates and performed chemical analyses on samples from 40 sites in urban areas of the United States from July 1979 to September 1981. 8711011 eg ------- |