Final Report, Volume III: Summaries of Individual AQAWG Analyses
OTAG Air Quality Analysis Workgroup
Dave Guinnup and Bob Collom, Co-chair
May 26, 1997
Table of Contents
SUMMARIES OF INDIVIDUAL AQAWG ANALYSES*
Spatial Pattern of Daily Maximum Ozone Over the OTAG Region*
Pattern of 8-Hour Daily Maximum Ozone Over the OTAG Domain*
Weekly Pattern of Ozone over the OTAG Region*
Spectral Decomposition of Ozone Time Series*
Ozone Exceedances Data Analysis: Representativeness of 1995*
Ozone Measurement Trend Studies in the Northeast*
SOS Nashville/Middle Tennessee Ozone Study*
Air Trajectory Analysis of Long-Term Ozone Climatology*
Source Regions of Influence for High and Low Ozone Conditions in the Eastern US*
Ozone/Tracer/NOY Relationships at Three SOS-SCION Sites*
Analysis of Ozone, NOY and Tracer Data from a Site in South-Central Pennsylvania*
A Comparison of Modeled and Measured Ozone, NOY and CO at Nine Regional Monitoring Stations during the 1995 OTAG Episode*
Ambient Monitoring Sites for OTAG (time series) Model Evaluation*
Comparison of OTAG UAM-V/BEIS2 Modeling Results with Ambient Isoprene Observations*
Comparison of SOS Nashville Data to OTAG 1995 Base Model*
Cover Image: a) Frequency of back trajectories for 22 receptor sites for low ozone days. These back trajectories point to outside the OTAG domain as the source of low ozone. b) Frequency of back trajectories for high ozone days (upper 50th percentiles). These back trajectories point to the OTAG domain as the source of high ozone.
SUMMARIES OF INDIVIDUAL AQAWG ANALYSES
A brief summary and critical review was prepared for most of the AQAWG analyses identified in Table 1 in Volume II: Summary and Integration of Results. These summaries provide a description of the purpose, methodology, and major findings of each study and are intended as a resource for quick reference in lieu of turning to the full study report(s). These summaries also include brief discussions of each study’s limitations and the scientific and policy implications for the issues being addressed by the OTAG. The summaries of findings, limitations, and implications were used as the basis for the integrated assessment presented above. Summaries are presented on the following pages in the order in which projects are listed in Table 1.
Spatial Pattern of Daily Maximum Ozone Over the OTAG Region
Participants: Rudy Husar, CAPITA, Washington University, St. Louis, MO
Reference: Spatial Pattern of Daily Maximum Ozone Over the OTAG Region, Rudy Husar 16 September 1996 (http://capita.wustl.edu/otag/Reports/otagspat/otagspat.html).
Purpose: Identify spatial patterns in daily maximum 1-hour ozone concentrations over the OTAG domain.
Method: A five year (1991-1995) data set of June-August daily maximum 1-hour ozone concentrations for the eastern U.S. was obtained by combining data from AIRS, CASTNET, and several other monitoring networks. All sites with at least 25 percent complete data were included in the analysis (thus, some of the sites used may only have data for less than a single season) . The final data consisted of a total of 715 monitoring sites. Six of these sites were discarded as a result of quality control procedures. Average concentrations and various percentiles of the concentration distribution were calculated for each monitoring site and contour plots prepared for each summary statistic. Specific emphasis was placed on the spatial distribution of the mean, 10, 50, and 90th percentiles.
Findings: Principal findings of this analysis were:
1. Mean daily maximum values range from a value close to 40 ppb (which is within the range of northern hemisphere tropospheric background values) in the "corners" of the OTAG domain (south Florida, south Texas Gulf Coast, upper Midwest/northern Plains, and northern Maine), up to about 80 ppb in the vicinity of some major urban centers. Mean values along the southern portion of the western boundary in Texas, Oklahoma, and Nebraska are about 10 ppb higher than those in the corners of the domain.
2. Mean daily maximum concentrations are highest in the Washington-New York urban corridor, with elevated concentration "hot spots" also observed in the vicinity of several other (but, surprisingly, not all) major urban centers.
3. A large region of elevated (65 - 70 ppb) mean daily maximum concentrations is evident along a band stretching across southern Indiana and Ohio, just north of the Ohio River.
4. East-west sections across the center of the domain show concentrations increasing consistently from west to east.
5. In the industrial Midwest, from Missouri through Illinois, Indiana, Ohio, Kentucky, and the Virginias, low (10 percentile) concentrations range from 42-47 ppb, about 15 ppb above values observed along the edges of the OTAG domain. However, in contrast to seasonal mean values, tenth percentile concentrations in urban areas are indistinguishable from their surroundings.
6. 90th percentile concentrations are highest in urban areas, including those urban areas which did not show mean concentrations elevated above regional background values. Elevated 90th percentile values are also found over a large region covering southern Indiana and southern Ohio.
7. Differences between 90th and 10th percentiles show that daily maximum ozone concentrations are most variable in urban areas and least variable in the corners of the OTAG domain. This is consistent with notion that monitored values in the corners of the domain represent broad-scale background ozone levels.
Limitations: These results are dependent on the spatial distributions of monitors, the spatial interpolation methods used, and the accuracy, precision, and completeness of the data. Quantitative estimates of spatial interpolation errors applicable to the AIRS/CASTNet data subset of this study are discussed by Falke (1996) but the impact of these errors are not discussed in connection with the principal findings of this study. A complete spatial analysis of interpolation errors for the integrated data set used in the spatial analysis would be needed to confirm some the findings noted above. In particular, since urban, near-urban, and rural sites were weighted equally in the interpolation procedure, it is possible that the interpolation from urban or near urban sites (which make up most of the network) produced a bias in the results in some locations. Also, since extreme concentrations (e.g., 90th percentiles) exhibit larger spatial gradients than mean or median values, the interpolation procedure may give a less accurate picture of the spatial pattern of extreme events as compared to average events. It should also be noted that the author uses a 25 percent data completeness criteria for including monitors in the analysis rather than the more commonly used 75 percent criteria. No analysis of the effect of the less stringent completeness criteria on uncertainties in computed summary statistics is provided. This is particularly of concern for the more extreme (i.e., 10 and 90) percentiles and the exceedance statistics. Application of the completeness criteria to the integrated five year data set rather than year-by-year is also a potential concern since missing values are unlikely to be randomly distributed over the five-year period and individual years are known to differ markedly in mean and peak ozone concentrations due to variations in meteorological conditions. Thus, blocks of missing values at key groups of monitoring sites could be biasing the results in unknown ways. Finally, it should also be noted that many or all of the days contributing to the low 10th percentile concentrations observed in many of the urban areas may have been impacted by NOx titration under conditions leading to high NOx concentrations and limited photochemical ozone production. This is in contrast to the authors conclusion that the low urban 10th percentile values reflect well ventilated conditions under which urban and surrounding rural concentrations are similar in magnitude. Additional analysis of the conditions contributing to the 10th percentile values will be needed to resolve this question.
Scientific Implications: Any realistic model of ozone/precursor relationships in the OTAG domain should be able to reproduce (or at least be consistent with) the principal features of the monitoring data described in this analysis. Of particular importance are the tropospheric background levels found in the corners of the domain and the apparent narrowing and shift to the right (i.e., towards higher concentrations) in the distribution of daily maximum values in southern Ohio and Indiana.
Policy Implications: Human activities must be responsible for some or most of the ozone excess in the central portions of the eastern U.S. over hemispheric background levels. Furthermore, these results suggest that precursor sources (presumably major NOx point sources) along the Ohio River Valley are associated with a broad region of elevated ozone levels not associated with any specific major urban centers. These results by themselves, however, do not indicate whether or not this broad region of elevated ozone concentrations is of concern from the stand point of limiting exceedances of the ambient ozone standard. Nor do these results by themselves indicate whether or not control of the major NOx sources geographically associated with this region will have a beneficial effect on reducing exceedances in the OTAG domain.
Pattern of 8-Hour Daily Maximum Ozone Over the OTAG Domain
Participants: Rudy Husar, Washington University, St. Louis, MO
Reference: Pattern of 8-Hour Daily Maximum Ozone and Comparison with the 1-Hour Standard. Rudy Husar, 16 September 1996 (http://capita.wustl.edu/otag/Reports/).
Purpose: Compare and contrast the spatial patterns of daily maximum 8-hour and 1-hour ozone concentrations over the OTAG region.
Method: A five year (1991-1995) data set of June-August hourly average ozone concentrations for the eastern U.S. was obtained by combining data from AIRS, CASTNET, and several other monitoring networks. All sites with at least 25 percent complete data were included in the analysis (thus, some of the sites used may only have data for less than a single season). The final data consisted of a total of 715 monitoring sites. Six of these sites were discarded as a result of quality control procedures. Daily maximum 1-hour and 8-hour averages were extracted for each day with sufficient data and means, 10, 50, 90th percentiles, and exceedance counts for various 1-hour and 8-hour thresholds were computed.
Results: Comparisons of daily maximum 1-hour values with 8-hour values for the same day indicated that the 8-hour to 1-hour ratios were lowest in urban areas where high peak 1-hour ozone concentrations are observed and approaches 1:1 at rural remote sites. For example, the average ratio at Greenwich, CT (adjacent to New York City) is 0.82 with ratios corresponding to daily maximum 1-hour values above 100 ppb averaging about 0.75. Other locations exhibit ratios in between these extremes. For the OTAG region as a whole, the average ratio is 0.86 with slightly higher values in the northwestern portion of the region and slightly lower values in the southwestern portion. Exceedances of a 1-hour, 120 ppb level are most common in the northeastern urban corridor and near other major urban centers. The spatial pattern of 8-hour exceedances of 102 ppb is similar. Moving to a 1-hour, 94 ppb or an 8-hour, 80 ppb level broadens the geographic extent of the exceedances to include a band of eight or more exceedance days per year just north of the Ohio river stretching from Illinois to Ohio.
and on into western Pennsylvania. Although the overall spatial patterns of 1-hour and 8-hour exceedances are similar, examination of a difference plot (8-hour exceedances of 80 ppb minus 1-hour exceedances of 94 ppb) indicates that the 8-hour exceedances are relatively more frequent within those broad areas north of the line separating Kentucky and Virginia from Tennessee and North Carolina which are away from the direct influence of a major urban area. With the exception of North Carolina, differences in exceedances are much less pronounced south of this line. Interestingly, Dallas and Houston, TX show quite different responses, with 8-hour exceedances relatively more common in Dallas and less common in Houston.
Maps of 90th and 10th percentile 8-hour daily maximum ozone concentrations are nearly identical to corresponding maps for the 1-hour daily maxima. Similarly, 8-hour and 1-hour maps of the difference between 90th and 10th percentiles are also very similar.
Limitations: Limitations noted for Husar’s analysis of 1-hour daily maxima also apply to this study with the exception that an appropriately more stringent data completeness criterion was applied to the analysis of 8-hour exceedances, thus reducing the potential for distortions resulting from missing data in these results. This study was done prior to EPA’s announcement of plans to replace the current 1-hour ozone NAAQS of 0.12 ppm with an 8-hour, 0.08 ppm standard and therefore did not directly compare areas of violation of the current and proposed standards, although exceedance maps for these two concentration levels are provided. Such area of violation maps have, however, been produced by EPA and others.
Scientific Implications: One-hour and eight-hour daily maximum concentrations are highly correlated both temporally and spatially. The spatial patterns of threshold exceedances and percentiles for 1-hour and 8-hour daily maxima are very similar to one another. Comparisons of the magnitude of same-day 1-hour and 8-hour peaks discussed above indicate that, under ozone episode conditions, concentrations at high ozone urban sites tend to exhibit relatively sharp 1-hour peaks. As a result, on a same day basis daily maximum 8-hour averages at such sites are relatively small compared to 1-hour maxima. In contrast, rural remote sites are characterized by low amplitude diurnal cycles in 1-hour values and therefore maximum 8-hour values tend to be only slightly less than corresponding 1-hour maxima.
Maps presented in this study show that multiple exceedances per year of an 8-hour, 80 ppb level covered a much broader area between 1991-1995 as compared to multiple exceedances per year of a 1-hour, 120 ppb level: the latter are confined to the immediate vicinity of a few large urban areas while the former cover much of the OTAG domain with the exception of the extreme northwest, northern Main, and the deep South indicating the broad spatial extent of such "mid-level" ozone concentrations. The threshold levels for which the frequency of 8-hour exceedances is comparable to the frequency of 1-hour exceedances of 120 ppb for the OTAG domain as a whole is approximately 102 ppb.
Policy Implications: Results presented in this study indicate that daily maximum 1-hour and daily maximum 8-hour concentrations are very closely related both spatially and temporally. Differences in the areas over which 1-hour vs. 8-hour standard violations occur depend on the level and form of the 8-hour standard and were not explicitly addressed here. However, the above results do indicate that at locations experiencing the highest 1-hour concentrations in the OTAG domain (e.g., downwind of major urban areas such as in southern Connecticut and along the eastern shore of Lake Michigan), the 1-hour peak exceeds the 8-hour peak by a greater amount than in other areas. Thus, all else being equal, an 8-hour standard may be relatively less stringent in these locations.
Weekly Pattern of Ozone over the OTAG Region
Participants: Rudy Husar, CAPITA, Washington University, St. Louis, MO
Reference: Weekly Pattern of Ozone Over the OTAG Region, Rudy Husar 16 September 1996 (http://capita.wustl.edu/otag/Reports/otagweek/otagweek.html).
Purpose: Analyze the influence of weekly cycles in anthropogenic emissions on daily maximum ozone concentrations.
Method: A five year (1991-1995) data set of June-August daily maximum 1-hour ozone concentrations for the eastern U.S. was obtained by combining data from AIRS, CASTNET, and several other monitoring networks. All sites with at least 25 percent complete data were included in the analysis (thus, some of the sites used may only have data for less than a single season) . The final data consisted of a total of 715 monitoring sites. Six of these sites were discarded as a result of quality control procedures. After disagregation by day of week, average concentrations and various percentiles of the concentration distribution were calculated for each monitoring site and contour plots prepared for each summary statistic. Specific emphasis was placed on the spatial distribution of the mean, 10, 50, and 90th percentiles and on exceedances of 120 ppb.
Results: For the OTAG region as a whole, seasonal peaks (90 and 95th percentiles) of daily maximum ozone concentrations were found to be about seven percent lower on Sundays than during the week. After correcting for a constant 40 ppb background, the weekday to Sunday difference is seen to represent about 33 percent of the OTAG region ozone excess above background. This "Sunday effect" is most pronounced in the 90th percentiles around the major urban areas where peak concentrations are highest but is notably absent (actually reversed with slightly higher values on Sundays than on Fridays) along a broad band stretching northeastward from western Tennessee through Kentucky, West Virginia near the Ohio border, Pittsburgh, and then northwestward to Cleveland. With the exception of Pittsburgh, this band overlaps the region of relatively high 10th percentile ozone concentrations noted by Husar (1996a). This "negative" Sunday effect is also evident at a few other isolated locations. No significant weekly cycle is evident in average or 50th percentile concentrations.
Weekly patterns in the frequency of exceedance of 120 ppb are similar to those for the 95th percentile of daily maximum concentrations with one third as many exceedances on Sundays as on Fridays. The number of exceedances increases nearly linearly from Monday through Friday. In most cases, the largest positive weekday - weekend differences in exceedances (i.e., more exceedances on Fridays than on Sundays) are over the urban areas with smaller differences elsewhere. In some urban areas, however, exceedances are actually somewhat more common on Sundays than on Fridays. Most of these urban areas fall within the geographic region noted above in which 90th percentile concentrations on Sundays are slightly higher than on Fridays. Several other isolated areas also exhibit a weak "negative" Sunday effect. The weekly cycle noted at the 120 ppb exceedance threshold is progressively damped at lower thresholds (100, 80, 60 ppb). Almost no weekend effect is evident at 60 ppb.
Scientific Implications: Results of this study indicate that the changes in average emissions between weekdays and weekends are sufficient to significantly reduce peak ozone concentrations on weekends while mean concentrations are largely unaffected. Unfortunately, since the nature (magnitude of emission reductions, relative source mix, diurnal pattern) of the differences in emissions on weekends as compared to weekdays have not been quantified, the mechanism resulting in lower peak ozone levels on weekends in urban areas is not yet fully understood. The gradual increase in upper percentiles of the concentration distribution and in the numbers of exceedances of high concentration thresholds (e.g., 120 ppb) during the course of the work week is indicative of the influence of carry over of ozone and precursors from one day to the next. This carry over keeps Saturday concentrations higher than they would otherwise be but by Sunday the full influence of the weekend emission changes are evident. The weekend effect is most pronounced at the upper end of the concentration distribution and therefore changes in urban areas, where high concentrations occur most frequently, tend to be larger than in rural areas. In some urban areas, however, peak concentrations on Sundays are actually slightly higher than on Fridays (e.g., Pittsburgh, PA). The reasons for this anomalous behavior are not known but may be related to geographic variations in combinations of several factors (mix of source types, activity patterns, VOC/NOx ratios). Geographically, the overlap between areas of relatively high 10th percentile concentrations and areas of negative weekend effect in the central portion of the OTAG domain may be related to the impacts of elevated NOx sources associated with base-load utility boilers characterized by relatively small weekly variations in emissions. However, the temporal pattern of emissions in this region has not been examined and the extent of any weekday/weekend differences are unknown.
Limitations: These results are highly dependent on the spatial distributions of monitors, the spatial interpolation methods used, and the accuracy, precision, and completeness of the data. Quantitative estimates of spatial interpolation errors applicable to the AIRS/CASTNet data subset of this study are discussed by Falke (1996) but the impact of these errors are not discussed in connection with the principal findings of this study. A complete spatial analysis of interpolation errors for the integrated data set used in the spatial analysis would be needed to confirm some the findings noted above. In particular, since urban, near-urban, and rural sites were weighted equally in the interpolation procedure, it is possible that the interpolation from urban or near urban sites (which make up most of the network) produced a bias in the results in some locations. Also, since extreme concentrations (e.g., 90th percentiles) exhibit larger spatial gradients than mean or median values, the interpolation procedure may give a less accurate picture of the spatial pattern of extreme events as compared to average events. It should also be noted that the author uses a 25 percent data completeness criteria for including monitors in the analysis rather than the more commonly used 75 percent criteria. No analysis of the effect of the less stringent completeness criteria on uncertainties in computed summary statistics is provided. This is particularly of concern for the more extreme (i.e., 10 and 90) percentiles and the exceedance statistics. Application of the completeness criteria to the integrated five year data set rather than year-by-year is also a potential concern since missing values are unlikely to be randomly distributed over the five-year period and individual years are known to differ markedly in mean and peak ozone concentrations due to variations in meteorological conditions. Thus, blocks of missing values at key groups of monitoring sites could be biasing the results. Finally, reasons for the "negative" weekend effect found in some areas are not well understood and require further study. In particular, no analysis is presented here of weekday/weekend differences in precursor emissions.
Policy Implications: Results of this analysis imply that a control scenario mimicking weekday to weekend emission reductions would result in decreases in peak ozone concentrations in most urban areas. For the OTAG region as a whole, exceedances of the current NAAQS (120 ppb) would be reduced by roughly a factor of three. Unfortunately, the exact nature of weekday/weekend emission differences is not currently known. Not only the magnitude of the emission reductions on weekends, but also changes in the relative mix of VOC vs. NOx emissions and of low-level vs. elevated NOx sources are likely to be important. Weekday/weekend differences in the diurnal pattern of emissions are also likely to be important: differences in the timing of mobile source emissions may be as important in reducing ozone concentrations as differences in the total amount of emissions.
Spectral Decomposition of Ozone Time Series
Participants: P.S. Porter, University of Idaho, Idaho Falls, ID 83405; S.T. Rao, I. Zurbenko, and E. Zalewsky, State University of New York, Albany, NY 12222; R.F. Henry and J.Y. Ku, NY State Department of Environmental Conservation, Albany, NY 12222.
References: Statistical Characteristics of Spectrally-Decomposed Ambient Ozone Time Series Data, P.S. Porter, S.T. Rao, I. Zurbenko, E. Zalewsky, R.F. Henry, and J.Y. Ku, Final Report, August, 1996.
Purpose: Estimate the spatial representativeness of data collected at routine ozone monitoring sites and the typical size of airsheds (zones of influence). Identify principal features of the spatial and temporal characteristics of daily maximum 1-hour and 8-hour ozone concentrations within three distinct frequency domains: long-term (i.e., annual), seasonal, and short-term (i.e., synoptic variations on time scales of less than about 30 days).
Methodology: Using multiple pass moving average filters (Kolmorgorov-Zurbenko, or KZ filters) described by Rao and Zurbenko (1994), the authors develop a spectral decomposition of the time series of natural logarithms of daily maximum 1-hour and 8-hour average ozone concentrations into long-term (annual), seasonal, and short-term (synoptic scale) components. This is done at all routine ozone monitoring sites in the U.S. reporting to EPA’s AIRS database for which few values are missing for the period 1983 - 1994. Monitors are segregated into those that operate year-round and those that operate only during the summer ozone season. Characteristics of each frequency domain are investigated, including temporal variance and autocorrelation, spatial patterns in means and variances, spatial correlations, and the influence of meteorological conditions.
Findings: Using two KZ filters, one a 15-day average, 5-pass filter and the other a 365-day, 3-pass filter, the authors find that the time series of log-transformed daily maximum concentrations can be cleanly separated into (1) a long-term component representing the influence of trends in emissions and (possibly) short-term climatic fluctuations, (2) a seasonal component representing the influence of the earth’s rotation about the sun, and (3) a short-term component representing the influence of fluctuating synoptic meteorological conditions and random processes (noise). An analysis of variance showed that covariances between the different components were generally small (less than 2% of the total variance). Long-term component variances were less than 10% of the total at more than ninety percent of all monitoring sites, with median values of 3.6% for summer season only monitors. For monitors with year-round data, the seasonal component accounts for up to 73% of the total variance (median value of 51%) while the median contribution of the short-term component is 51%. For summer season only monitors, the median seasonal component variance contribution was 12%, while the median short-term component was 77%. Seasonal components are higher in the Northeast than in the South. Contributions of short-term variance to the total are highest in coastal areas and lowest in the Midwest. In absolute terms, the highest short-term variances are in the Northeast and along the Gulf Coast.
An evaluation of the statistical properties of the short-term component, W(t) revealed it to be significantly negatively skewed and therefore not normally distributed at nearly all monitoring sites, while exp(W) was more nearly normally distributed in most locations. Autocorrelations in W(t) at one day lag had a median value of 0.35, indicating some day-to-day dependence resulting from persistence in meteorological conditions. Nevertheless, modeling the short-term component as a random variable normally distributed with zero mean, the authors found that it is possible to reproduce the observed median concentrations at nearly all sites (to within the 95% confidence interval of the predicted value); 95th percentiles of the daily maximum values were less accurately reproduced. Exceedance counts were similarly evaluated, but it is not clear how to interpret these results since most monitors will have had very few or no exceedances over the 12-year period examined. Rao et al. (1996) provide comparisons of observed and modeled exceedance counts and 95th percentiles at five locations which experienced multiple exceedances per year (at least prior to 1989) which show generally close agreement with some exceptions.
A trends analysis for the period 1985 - 1995 was also presented by the authors. Trends were computed for the ozone baseline (long-term plus seasonal component) both before and after adjusting for variations in temperature. Raw (unadjusted) baseline trends are mixed, with generally negative trends in the southern tier states, Illinois, and the Philadelphia - New York urban corridor but increases at most sites in Indiana, Ohio, West Virginia, Western Pennsylvania, and the Washington D.C. area. After adjusting for temperature fluctuations, negative trends were found at a much larger number of sites.
Spatial correlations in baseline and short-term components of both ozone and surface temperature time series were examined. For the Washington D.C. area, temperature baseline correlations remained constant and near unity over the maximum distance examined (1,280 km or 800 miles), while correlation in the short-term component decreases linearly with distance over this radius, dropping to 0.3 at 1,120 km (700 miles). Ozone baseline spatial correlations drop off more rapidly but also respond approximately linearly with distance, at least out to the maximum distance considered (800 km or 500 miles), at which point the correlation coefficient is approximately 0.8. Short-term ozone component spatial correlations decrease exponentially with distance over this range, dropping to a value of 0.3 at about 720 km (450 miles) along the "direction of transport". A map of spatial correlation isopleths for the short-term component is presented by the authors for five cities in the OTAG domain (Atlanta, Chicago, Cincinnati, Greenbelt MD, Pittsburgh). In all cases, correlations drop to 0.4 within 560 to 640 km (350 to 400 miles) of the selected monitors. It is suggested that these are the spatial domains which should be considered in designing control strategies.
Comparisons of time series of 1-hour average daily maximum ozone to 8-hour average daily maximums indicated that variances are higher for the 8-hour averages, both in an absolute sense, relative to the mean, and for the short-term component only. Looking at daily maximum n-hour averages at one site (Cliffside Park, NJ), the authors found that variances of the short-term components peak at n=10 hours (s = 0.38 as compared to s = 0.27 at n=1 and s = 0.35 at n=24). The statistical significance of these differences are not discussed.
Finally, using the technique described above of modeling the short-term component, W(t) as white noise, the authors compute the percent reduction needed in the ozone baseline component to achieve attainment of the ozone NAAQS for the 1987-1989 and 1991-1993 attainment periods at each monitoring site. These required reductions range from 0% (for monitors already in attainment), to 50% at monitors in high ozone locations in the Northeast. Generally lower reductions are required during the latter period due to the progress towards attainment already achieved.
Limitations: From a regulatory perspective, much of the value in these analyses is based on the viewpoint that the baseline component of the ozone time series (i.e., the component remaining after removal of short-term fluctuations using the KZ15,5 filter) is "deterministic" in the sense that it is controlled entirely by seasonal meteorological fluctuations and year-to-year trends in emissions. The remaining, short-term component, on the other hand, resemble random fluctuations. Thus, management of the ozone problem should focus on the effects of control strategies on the baseline rather than being concerned with a highly fluctuating peak statistic (e.g., 4th highest concentration in three years) in which much of the variability is associated with "noisy" processes that cannot be accurately modeled. However, in order to make use of this approach, one must be able to accurately estimate the impact of future control strategies on the baseline, and such methods have yet to be successfully demonstrated (i.e., it is not clear how to adapt episodic photochemical models to models of baseline ozone). In addition, one must be able to estimate the impact (or convincingly demonstrate that there is no impact) of control scenarios on the statistical characteristics of the short-term component. To date, neither of these conditions have been met. Of course, it should also be noted that the ability of episodic photochemical models to accurately predict the impact of future-year control strategies on ozone levels has also yet to be successfully demonstrated.
Although the trends analysis results presented by the authors are in general agreement with results of other studies (see review prepared by Morris, 1996), the fact that nearby sites often exhibit opposing trends is indicative of a complex spatial structure in trend statistics which may not be fully resolvable with the current monitoring network. Due to the large number of sites examined, the trend maps presented by the authors are difficult to read in urban areas with many closely spaced monitoring sites due to the overlapping of green disks (denoting negative trends) on top of red ones (denoting positive trends); green disks are always on top, thus obscuring an unknown number of sites with positive trends. Tables of trend values for each site are available from the authors for review and Rao et al. (1995) present a table of trends for sites in the eastern U.S. for 1983 - 1992. Similar trend maps for the northeastern U.S., but showing spatially smoothed trends (where the smoothing was done with spatial KZ filters) are presented by some of the same authors (Zurbenko, et al., 1995) and these maps do not suffer from this difficulty. Unfortunately, the spatial smoothing may obscure some important small scale trend differences. Also, interpretation of trend statistics would be enhanced by further characterization of trends by site type (i.e., rural, remote, suburban, downwind urban peak, urban center, etc.).
Published demonstrations of the independence of short-term components of daily maximum ozone and temperature (Rao and Zurbenko, 1994; Flaum, Rao, and Zurbenko, 1996) are not conclusive because seasonal influences on the correlation were not resolved. Therefore, it has not been demonstrated that the short-term ozone fluctuations during the summer season are completely independent of meteorological effects. Thus, spatial correlations in this component may be partially attributable to spatial correlations in meteorological conditions in addition to emissions forcing and defining a spatial domain for control strategy development on the basis of zero lag time spatial correlations in the short-term components may not properly address the area of source influence. Furthermore, correlations at non-zero lag times (e.g., one or two day lags) also need to be considered. Spatial scales of correlations in the short-term temperature component appear to be reasonable from a synoptic scale meteorological perspective. The fact that these scales are greater than those for the ozone component may indicate that much of the remaining variance in the ozone component (i.e., that portion not related to temperature) is associated with other meteorological factors that operate on smaller spatial scales.
Implications: Perhaps the most important implication of the above findings is that, when only the summer high ozone season is considered, short-term fluctuations are the dominant source of variability in daily maximum concentrations (either 1-hour or 8-hour average). Thus, identifying the impact of control strategies on ozone levels is extremely difficult due to the low signal to noise ratio. Furthermore, areas close to the attainment /nonattainment threshold are likely to jump in and out of attainment (where attainment/nonattainment is determined on the basis of one to three years of monitoring data), thus complicating management efforts. This study shows that the problem appears to be particularly severe in the Northeast - precisely where accurate estimates of control strategy impacts are most needed.
When seasonal and long-term trends are removed from the ozone time series, the above results show the spatial scale of the remaining short-term fluctuations to be 560 - 640 km (350 - 400 miles), suggesting a scale for the coherence of same-day fluctuations in peak ozone which may be useful for air quality management efforts. However, the degree to which this spatial scale is determined by spatial correlations in meteorological conditions as compared to same-day transport of ozone and precursors is unknown.
Morris, 1996. Review of Recent Ozone Measurement and Modeling Studies in the Eastern United States. ENVIRON International Corp., 21 February 1996.
Rao and Zurbenko, 1994. Detecting and Tracking Changes in Ozone Air Quality. J. Air and Waste Management Assoc., Vol. 44: 1089 - 1092, 1994.
Rao, S.T., Zurbenko, I.G., Porter, P.S., Ku, J.Y. and R.F. Henry, 1996. Dealing with the Ozone Non-Attainment Problem in the Eastern United States. Environmental Manager, January, 1996, pp. 17-31.
Flaum, J.B., Rao, S.T., and I.G. Zurbenko, 1996. Moderating the influence of meteorology on ambient ozone concentrations. J. Air & Waste Manage. Assoc., 46: 35-46, 1996.
Zurbenko I.G., Rao, S.T. and R.F. Henry, 1995. Mapping Ozone in the Eastern United States. Environmental Manager, Vol. 1 pp.24-30, January, 1996.
Ozone Exceedances Data Analysis: Representativeness of 1995
Participants: Lyle Chinkin, Richard Reiss, Douglas Eisenger, Timothy Dye, and Christopher Jones, Sonoma Technology, Inc., Santa Rosa, CA
Reference: Ozone Exceedances Data Analysis: Representativeness of 1995. L. Chinkin et al., Sonoma Technology, Inc., 1996.
Purpose: Assess the representativeness of the 1995 ozone season in the Northeast in the context of the past 15 ozone seasons. The motivation for this study was the realization that the summer of 1995 was one of the hottest on record for the Northeastern United States, and that the 1995 summer season was therefore a strong candidate to experience high ozone concentrations.
Methodology: The authors undertook an investigation of whether the above normal temperatures during the summer of 1995 were associated with above normal ozone concentrations. The authors utilized air quality and meteorological data from seven metropolitan areas within the Northeast’s Ozone Transport Region (OTR), as well as the OTR overall. The metropolitan areas studied include Baltimore, Boston, Hartford, New York, Philadelphia, Providence, and Washington, D.C. Numerous regional and subregional trends in 1-hour average ozone concentrations were evaluated for the 1980s and 1990s to determine to what extent 1995 was consistent with these trends. Year-to-year fluctuations in the frequency and severity of meteorological conditions conducive to ozone formation were accounted for using several different methods including: 1) An examination of year-to-year differences in the relationship between the frequency or severity of high temperature days (e.g., the number of days above 90°F) and the frequency or severity of ozone episodes (e.g., number of exceedance days), 2) the Poison regression model developed by Cox and Chu (1995), and 3) The synoptic typing scheme of Comrie and Yarnal (1992).
Findings: It was found that despite near record breaking temperatures, the 1995 ozone season was no worse than the average ozone season experienced during the 1990s. In fact, 1995 is consistent with overall trends showing declining ozone concentrations in the Northeast. The major findings cited by the authors are as follows:
Limitations: As with most other trends studies, the emphasis here is on peak concentrations in urban areas in the Northeast. However, the study did include a comparison of trends at sites designated as urban or suburban in the AIRS database with sites determined to be in rural locations based on a review of their map coordinates. This analysis indicated no discernable difference between urban and rural trends. It should also be noted that uncertainties in the emission inventory trends (especially in the NOx trends) are not evaluated. This is particularly important given the lack of attention to ambient NOx trends which could be used to corroborate the inventory trend (assuming sufficient high quality NOx data are available). Uncertainties in the correction for meteorological conditions must also be taken into consideration. In particular, the Cox and Chu Poison regression method assumes underlying (i.e., emission-related) trends are linear in time - this is a rough approximation at best. In addition, year-to-year changes in the ratio of the number of days in a given season above a temperature threshold to the number of ozone exceedance days are very sensitive to the temperature and concentration thresholds used and are subject to numerous factors not related to emissions. These ratios thus provide only a very rough indication of actual precursor emission trends.
Scientific Implications: Results of this study are generally consistent with those of other recent trends analyses which show concurrent declines in peak ozone levels and VOC emissions. Data on ambient precursor concentration trends are needed to verify estimated trends in emissions.
Policy Implications: Although not entirely definitive for the reasons noted above, this study supports other independent analyses of northeastern ozone and emission trends, strongly suggesting that recent VOC emission reductions resulted in decreases in peak 1-hour ozone levels over the routine monitoring network. Concurrent but relatively small NOx emission reductions may have contributed to the ozone reductions. Of course, these studies do not provide any direct information on the relative effectiveness of future VOC controls in the region.
Comrie, A.C. and B. Yarnal, 1992. Relationships between synoptic-scale atmospheric circulation and ozone concentrations in metropolitan Pittsburgh, Pennsylvania. Atmos. Environ., 26B, 301-312.
Cox, W.M. and S-H Chu, 1996. Assessment of interannual ozone variations in urban areas from a climatological perspective. Atmospheric Environment, 30(14): 2615-.
Ozone Measurement Trend Studies in the Northeast
Participants: Ralph Morris, ENVIRON, 101 Rowland Way, Novato, CA (reviewing work performed by USEPA (1994 Air Quality Trends Report), S. T. Rao et al., New York State Department of Environmental Conservation, George Wolff and Patricia Korsog, General Motors Corporation (GMC), William Cox and Shao-Hang Chu, OAQPS, USEPA, Kay Jones, Zephyr Consulting
Reference: Review of Recent Ozone Measurement and Modeling Studies in the Eastern United States. R.E. Morris, ENVIRON International Corp., March, 1996 and references therein.
Purpose: Review and prepare an integrated summary of recent major northeastern U.S. air quality and emission trend analyses.
Methodology: Several researchers have performed trends analysis of ozone in the eastern U.S. in recent years. These analyses differ mainly in the techniques used to account for the impact of meteorological variations in the ozone trends. A brief summary of methodologies and results were prepared for each analysis. These summaries were then used as the basis for an integrated assessment of recent ozone and emission trends in the Northeast.
Findings: Summaries of methods and results for each individual study:
EPA National Trends Reports: EPA’s 1994 annual trends report (EPA, 1994) examined both longer-term (1984-1993) and shorter-term (1991-1993) trends in ozone using national composite averages (EPA’s 1995 trends reports was not available). Between 1984 and 1993, the composite average exceedances over the 532 monitoring sites examined decreased 60 percent, a statistically significant amount. The EPA national trends report also examined the trends in ozone concentrations after the effects of meteorological variations have been removed using the parametric regression technique developed by Cox and Chu (1992). The meteorological-adjusted ozone concentrations illustrated a definite downward trend in ozone over the period studied. EPA’s 1995 emissions trends report (EPA, 1995) provides trends in national ozone precursor emissions over the period of 1900-1994, with particular focus on the last ten-year period of 1985-1994. During this ten-year period, there was a distinct downward trend in VOC emissions (14 percent reduction) and a slight increase in NOX emissions (6 percent increase) suggesting that the downward trend in ozone is likely due to the reductions in VOC emissions.
Accounting for Meteorological Variations in Ozone Trends by NYSDEC/SUNY: Researchers from the State University of New York (SUNY) at Albany and the New York Department of Environmental Conservation (NYSDEC) have developed a statistical technique to eliminate the high frequency variations in ozone trends due to meteorological fluctuations (Rao and Zurbenko, 1994). Rao and co-workers applied this technique to analyze the trends in meteorological-adjusted ozone concentrations for the period of 1980-1992. They found a definite downward trend in ozone for the Northeast Corridor stretching from Wilmington, Delaware up to Boston, Massachusetts, and essentially no change in the meteorological-adjusted ozone in the Baltimore-Washington D.C. and southern Maine regions. Zalewsky and co-workers (1994) examined the trends in VOC and NOX emissions in the Northeast Corridor during this period and found a downward trend in VOC emissions and very little change in NOX emissions. This suggests that the decrease in ozone concentrations are attributable to the VOC emissions reductions.
General Motors Corporation Studies on Ozone Trends: A series of studies performed by researchers at General Motors Corporation (GMC) analyzed ozone trends at several cities in the eastern U.S. for the period of 1980 through 1993, as well as trends in morning VOC-to-NOX ratios and the Reid Vapor Pressure (RVP) of gasoline as a surrogate for VOC emissions (Wolff, 1993; Wolff and Korsog, 1994). Although the GMC researchers did not perform adjustments to the ozone trends to account for meteorological variations, trends were calculated for several different ozone parameters, including 8-hour averages and number of days the maximum 1-hour and 8-hour ozone exceeds threshold levels. They also calculated confidence intervals to determine whether trends were statistically significant. They noted statistically significant downward trends in many of the ozone summary statistics in New York, Newark, Milwaukee, and Chicago, and downward trends in maximum and second highest 1-hour ozone, although not always statistically significant, in all cities studied except Boston and Grand Rapids. It is interesting to note that in Atlanta and Philadelphia (and to a lesser extent in Baltimore), there were downward trends in peak 1-hour ozone concentrations, but upward trends in the number of days in which the 8-hour ozone concentration exceeded 60 ppb. These results suggest that the distribution in --hour ozone concentrations in these cities is becoming narrower and urban peaks are declining, but mid-level (baseline) ozone concentrations are increasing. The GMC researchers’ analysis of the trends in VOC-to-NOX ratios also suggest a downward trend in the cities studied from a value of approximately 10 in 1986 to around 5-6 in 1991. These results suggest that urban photochemistry is becoming more VOC-limited in these cities (i.e., further VOC emission reductions will be effective in lowering ozone concentrations, while NOx emission reductions may result in local increases in ozone concentrations).
Trends in Meteorological-Adjusted Ozone Concentrations using the Cox and Chu Method: Cox and Chu (1993, 1995) used a probability model to examine meteorological-adjusted ozone trends with statistical confidence estimates. For the period of 1981 to 1990, statistically significant declines in ozone concentrations were calculated in New York, Philadelphia, Pittsburgh, and Baltimore, and small, but not statistically significant, declines in Washington D.C. and Boston were calculated. Cox and Chu also used their probability model to estimate the relative potential for the occurrence of meteorological conditions conducive for ozone formation in each urban area for the 40 year record of 1953 to 1993. For northeastern urban regions, the rank order statistics suggest that 1988 had the most conducive meteorological conditions for ozone formation in the 40 year record followed by 1953, 1955, 1983, 1991, and 1993. The return time of the extremely adverse conditions seen in 1988 was estimated to be a 1-in-40 year event in most northeastern cities.
Meteorological-Adjusted Trends in Ozone Concentrations Calculated by Kay Jones: Kay Jones of Zephyr Consulting performed several analyses of trends in yearly maximum, second highest, and three-year design value (fourth highest ozone value in three years) ozone concentrations. Year-to-year variability in meteorology were accounted for by the number of days per year the temperature in Philadelphia exceeds 90 F. A steep decline in meteorological-adjusted ozone concentrations in all northeastern cities was estimated. Using more recent ozone data to calculate design values suggests there would be fewer ozone nonattainment areas than the three-year period that was used to classify them in the 1990 CAAA (which included 1988).
Integrated Results: According to the author’s review, the measurement studies demonstrate that: (1) meteorological conditions play an important role in the occurrence of ozone exceedances; and (2) after accounting for meteorological variations, there is a definite downward trend in ozone concentrations over the last ten years in the eastern U.S. Because VOC emissions have been reduced during this period while NOX emissions have remained fairly stable, this air quality improvement can be attributed to reductions in VOC emissions. Although most of the studies focused on 1-hour ozone concentrations, one study also considered 8-hour average ozone concentrations. This study found that while the peak 1-hour concentrations decreased, the 8-hour concentrations increased. This may indicate either that the diurnal variation in ozone is getting narrower or that the VOC reductions are effectively reducing the urban peak 1-hour ozone concentrations, but they have less effect on regional ozone (which is more closely represented by the 8-hour ozone concentration).
Limitations: The principal focus of this review was on modeling studies; thus the review of measurement studies was not comprehensive. Furthermore, most measurement programs are focussed on nonattainment areas, thus the trends are biased towards urban areas; trends in rural areas are not specifically analyzed. It should also be noted that uncertainties in the emission inventory trends (especially in the NOx trends) are not evaluated. This is particularly important given the lack of attention to ambient NOx trends which could be used to corroborate the inventory trend (assuming sufficient high quality NOx data are available).
Scientific Implications: Recent declines in (predominantly urban) peak 1-hour ozone levels in the northeastern U.S. (after at least partially accounting for variations in meteorological conditions) are coincident with decreases in VOC emissions. In contrast, inventory NOx emissions are relatively stable during this period. In at least one case, opposing trends in 8-hour vs. 1-hour concentrations have been noted, implying either that VOC controls may be decreasing peak concentrations or that diurnal variations in ozone are getting narrower. Data on ambient precursor concentration trends are needed to verify estimated trends in emissions. Additional information on rural area trends is also needed.
Policy Implications: Although not entirely definitive for the reasons noted above, these independent analyses of northeastern ozone and emission trends strongly suggest that recent VOC emission reductions resulted in statistically significant decreases in peak 1-hour ozone levels over the routine monitoring network. Of course, these studies do not rovide any direct information on the relative effectiveness of future VOC controls in the region.
SOS Nashville/Middle Tennessee Ozone Study
Participants: Jim Meagher, Tennessee Valley Authority
References: Slides from presentation given to the OTAG AQAWG meeting, Washington D.C., July 1996.
Purpose: Analyze ambient data from the 1995 Nashville SOS study to investigate ozone formation in the Nashville area and in major point source plumes released into a southern environment.
Methodology: The SOS Nashville data base for 6 weeks in June and July, 1995, includes enhanced measurements of air quality and meteorological parameters, e.g., surface/aloft VOC, CO, SO2, NOX and NOY concentration data, airborne LIDAR ozone data, ozone sondes, and RADAR/RASS atmospheric profilers. There were over 200 scientists involved, and Nashville 1995 was one of the largest air quality studies ever conducted in the U.S. This period overlaps with the July 9-18, 1995 OTAG episode. The study also covered a four week period in 1994, but aircraft measurements were much more limited in 1994 than 1995. For 1995, surface and aircraft ozone and NOY data were analyzed for ozone productivity, defined as the ratio of O3 to NOZ (where NOZ = NOY - NOX) which may be considered as the number of ozone molecules produced per NOX molecule lost (ozone productivity analyses are also discussed in the review of presentations by Bob Imhoff of TVA). Surface and aloft ozone data were analyzed to characterize the spatial extent (horizontal and vertical) of the Nashville ozone episode that occurred over July 11-15, 1995. Aircraft data collected in several large point source NOX plumes were analyzed to characterize the spatial/temporal scales for conversion of NOX to NOZ and the concomitant ozone production.
Findings: The Nashville ozone episode that occurred over July 11-15, 1995 appears to be "home grown," i.e., the area of high ozone was quite localized around Nashville and therefore appears to have been related to Nashville emissions. Ozone concentrations in the mixed layer around Nashville appear to have been elevated by up to about 80 ppb above the regional ozone concentration of 60-70 ppb. The mixed layer depth extended up to about 2 km and ozone concentrations above the mixed layer were about 50 ppb (aircraft data).
The author also compared O3 to NOZ relationships in urban and power plant plumes using two approaches. In one method, O3/NOZ relationships measured at three ground stations were compared for periods of high and low SO2 levels, where the SO2 levels were used to indicate whether the airmass was associated with power plant or urban emissions. The ozone production efficiency was 70% higher for air masses with low SO2 levels, possibly indicating that ozone production is more efficient (in terms of NOX utilization) in urban plumes than in power plant plumes. O3/NOZ relationships were also evaluated from aircraft data obtained in urban and power plant plumes. On one day, July 16, the TVA helicopter made measurements in both a power plant plume and an urban plume and observed different ozone production efficiencies, with the urban plume approximately 65% more efficient than the power plant plume.
Plumes from the Cumberland (~350 tons per day of NOX) and Johnsonville (~80 tons per day of NOX) power plants were followed by helicopter flights from close to the source out to about 150 km downwind. At this distance, NOX was observed to be essentially totally converted to NOZ, and NOX levels had dropped to a point where ozone production was essentially over (this situation is characterized as one where the NOX reactivity is exhausted). For the Johnsonville plume on June 23, ozone levels were enhanced by up to ~25 ppb over a background of ~75 ppb at the point where NOX reactivity was exhausted. This plume filled the mixed layer and plume conditions were sampled at an altitude of about 1200 m M.S.L. Based on the plumes followed in this field study, the authors concluded that NOX reactivity is exhausted (and ozone formation is over) within 30 to 100 km downwind.
Limitations: The SOS Nashville data base appears to be rich and robust, so that limitations of these findings are likely to be related the number of events captured during the study (4 weeks in 1994 and 6 weeks in 1995). The July 11-15, 1995 Nashville ozone episode is only one event and may not be representative of all Nashville ozone episodes, thus not all Nashville episodes may be "home grown." The conclusions regarding the lifetime of NOX reactivity in power plant plumes are for daytime meteorological conditions typical of the Nashville area in the summer, and may not be applicable to other regions and/or times of year. Also, power plant plumes released at night are generally subject to less turbulent mixing and may remain intact longer. These NOx plumes may thus travel relatively long distances before taking part in photochemical reactions the following day.
Scientific Implications: For the July 11-15, 1995 Nashville ozone episode, enhancement of ozone above background due to local emissions appeared to dominate over regional ozone enhancement. Of course, the characteristics of this episode may differ from other episodes in the region. The reactivity of NOX in power plant plumes released into the southern, daytime summer boundary layer appears to be rather short-lived (30 to 100 km downwind). Also, these observations suggest that ozone production is more efficient (in terms of NOX utilization) in urban plumes than in power plant plumes
Policy Implications: If Nashville is representative of other urban areas in the south, and if the July 11-15, 1995 episode which was the focus of the results described here is typical of ozone episodes in these areas, then the results presented here suggest that effective ozone control measures should focus on reducing local urban emissions. The short-lived reactivity of NOX in power plant plumes found in this study suggests that during the day, the ozone formation generated by the plume is completed within about 100 km of the source. On the other hand, plumes released at night may travel relatively longer distances before taking part in ozone generating reactions the following day.
Air Trajectory Analysis of Long-Term Ozone Climatology
Participants: Rich Poirot and Paul Wishinski, Vermont Department of Environmental Conservation.
VT DEC Air Trajectory Analysis of Long-Term Ozone Climatology: Status Report to OTAG Air Quality Analysis Workgroup, 12/3/96. R. Poirot and P. Wishinski, 3 December 1996
VT DEC Air Trajectory Analysis of Long-Term Ozone Climatology: Status Report to OTAG Air Quality Analysis Workgroup, 11/7/96. R. Poirot and P. Wishinski, 7 November 1996
VT DEC Air Trajectory Analysis of Long-Term Ozone Climatology. Status Report to OTAG AQA Workgroup: 8/15/96. R. Poirot and P. Wishinski, 15 August, 1996 (http://capita.wustl.edu/otag/Reports/vtdecair/vtdecair.html)
Air Trajectory Residence Time Analysis Investigation of Ozone Transport Pathways: 1989-95. P. Wishinski and R. Poirot, 20 February, 1996 Summary Draft (http://capita.wustl.edu/otag/Reports/Restime/Restime.html)
Purpose: Perform a back-trajectory based analysis of relationships between high regional ozone levels and transport pathways. Develop trajectory climatologies for regionally representative receptor locations throughout the OTAG domain and use trajectory sorting techniques to identify predominant synoptic-scale transport pathways when ozone is high at the receptors.
Methodology: The basis for the approach is calculation of trajectories running backwards in time from a selected receptor/time. Multiple trajectories for the same receptor but different end-times (e.g., 9 a.m. on all summer days) are assemble into families that define a trajectory climatology. The trajectory climatology suggests where air arriving at the receptor within the selected family of end times had previously resided. A technique known as "residence time analysis" is used to display integrated results for large numbers of back trajectories. The approach is to grid the trajectory domain, keep track of the time each trajectory spends over each grid square along its path, and integrate the residence times for all trajectories over all grid squares. The resulting grid of residence times are displayed as a gridded density plot called a residence time plot.
Trajectories ending at a given receptor can be sorted and grouped by various criteria. For concentration based sorting, trajectories are sorted on the basis of the ozone concentration occurring at the time of their arrival at the receptor (e.g., above average, above 60 ppb). The resulting residence time plot addresses the question "if the concentration at this site was high (or low or very high), where did the air come from?" Residence time probabilities are computed for each grid square as the ratio of the total residence time in the square from a group of trajectories (e.g., those corresponding to above average ozone at their end points) to the total residence time over all grid squares for the trajectory group. Such probabilities tend to be highest in the grid squares closest to the receptor since all (or nearly all) trajectories must pass through these grid squares regardless of the end point ozone concentration. However, by subtracting the residence time probability for a selected group of trajectories from the residence time probability for all trajectories, a gridded density plot of the incremental residence time probability is obtained. Incremental residence time probability plots indicate the areas which are relatively more likely to be upwind under the selected set of conditions and thus eliminate the "near field" bias.
A different type of sorting is location based sorting. Here, an average (or median) ozone level is calculated for each grid square as the weighted average concentration at the end points of all trajectories that passed through the grid square. Trajectories are weighted according to their residence times in the grid square. Resulting plots display average concentrations at the receptor for prior airmass locations on the grid and address the question "if the air had previously been here (or there), what’s the average concentration at the receptor?" Typically, these average concentrations are expressed in terms of normalized deviations from the receptor mean concentration. Location based sorting was used to draw maps showing the locations of grid squares where the residence time weighted average end point concentrations were above or below the average receptor concentration by varying amounts.
Back-trajectories were calculated using the NOAA HY-SPLIT model with archived wind fields from the NGM weather forecasting model (Draxler, 1992; Phillips, 1975). Trajectory end times examined were 3 am, 9 am, 3 pm, and 9 pm for June-August 1989-95. Results are presented for six high elevation sites in the Appalachian Mountain chain from Look Rock, TN to Whiteface Mountain, NY. Results are also presented for 19 low elevation sites in an area ranging (approximately) from the Gulf Coast to the Canadian border and the Mississippi River to the Atlantic Coast.
In most of the analyses, raw ozone data were transformed to "Z-scores", based on geometric mean and standard deviations - computed separately for each site, for each hour of day. All resultant Z-score distributions have a mean of 0 and standard deviation of 1. This standardization technique was used to allow data from different hours of the day and from different sites to be merged and sorted into consistent subsets of the statistical distribution (i.e. upper 50th, 10th or 5th percentiles). Z-scores were used to define various subsets of trajectories and were also employed in a variety of inter-site correlations, time-lagged auto-correlations, and time-lagged inter-regional correlations - for comparison to the trajectory results. For the inter-regional analyses, six subregions within the OTAG domain (New England, Mid-Atlantic, Northwest, Midwest, Southeast, and Southwest) were defined on the basis of groups of monitoring sites in close geographic proximity with high inter-site correlations.
Findings: For the complete trajectory ensemble (i.e. unsorted), eastern receptors such as Whiteface Mountain (NY) and Look Rock (TN) have 2 to 4 day back trajectories covering almost the entire OTAG domain, suggesting the potential for transport of air parcels to these high elevation sites from across the whole OTAG region. Residence time analyses of trajectories ending at both the high and low elevation sites sorted for high end point ozone levels (based on normalized deviations from the receptor site means) suggest high ozone is generally associated with transport from the heart of the OTAG domain rather than from around the edges. Similarly, location based sorting of trajectories indicates that trajectories which traversed areas within the central portion of the OTAG domain produced higher end point ozone concentrations on average, regardless of the location of the receptor. The incremental probability plots for many receptors or groups of receptors show that, as the analysis is increasingly restricted to the highest ozone concentrations, the area of highest incremental probability becomes increasingly focussed on an area around the Ohio River Valley. Within the southern portion of the OTAG domain, trajectories associated with above average ozone tend to extend a smaller distance from the receptor than in the northern portion of the domain, indicating a relatively greater influence of stagnation in the South.
The finding of relatively more stagnation in the south is supported by an analysis of the autocorrelation in ozone concentrations (mean 3:00 pm Z-scores) within each of the six subregions within the OTAG domain: autocorrelations at one-day lag times were above 0.4 in all subregions, while autocorrelations remained well above 0.2 even at a three-day lagtime in the Southwest and Southeast subregions. In contrast, autocorrelations dropped below 0.1 at two-day lagtimes in the New England, Mid-Atlantic, and Northwest subregions.
Limitations: Residence time analysis of trajectory climatologies can suggest regional airflow patterns associated with high ozone regimes at a receptor, but without information on the age of the elevated ozone reaching the receptor, inferring the location (i.e., distance upwind) of the emissions that caused the elevated ozone is problematic; the trajectory results do not account for chemical transformation or deposition processes. These problems apply to all trajectories but become particularly acute for trajectories extending back more than about 12 hours. Location based sorting seems as if it may address this issue, but it is difficult to know how to interpret the plots of location sorted trajectories. One difficulty is that location plots sorted for high and low ozone frequently show no contribution from grid squares immediately upwind of the receptor. These squares tend to be classified as average because many trajectories of both high and low ozone pass through them The effects of this ‘far-sighted’ bias in the location based sorting have not been estimated. These limitations imply that relationships between the patterns of trajectory residence times and end point ozone concentrations do not necessarily indicate a cause and effect relationship between emissions in a given area and end point ozone concentrations - these relationships may simply be an expression of the prevailing meteorological conditions under which high ozone concentrations at the end point receptor are favored.
Because all the analyses are limited to regionally representative receptors, the high ozone levels observed are often below the ozone NAAQS. This fulfills the objective of identifying transport conditions for high regional ozone levels, but this study does not specifically address how such conditions relate to transport conditions during ozone NAAQS exceedances. It must also be recognized that these analyses are limited by uncertainties and/or biases in the back trajectory calculations, and these uncertainties can be expected to grow as the trajectories are computed further back in time. Trajectories computed from NGM wind fields with relatively coarse vertical resolution are not likely to reflect local or mesoscale flow features such as low-level nocturnal jets, sea breezes, and other small scale flow features likely to be important to ozone and precursor transport under ozone episode conditions. Some trajectory uncertainties have been investigated by Schichtel and Wishinski (1996) by comparing alternate trajectory models, and these uncertainties do not appear to exert a major influence on the results presented here. Although some trajectories are calculated for as long as four days at which time the uncertainties are relatively large, most trajectories are truncated at 2 to 3 days for various reasons. Furthermore, the authors indicate that they are unaware of any sources of bias in the analyses (nor were any biases found by Schichtel and Wishinski) and therefore that results averaged over a large number of trajectories calculated under a particular set of conditions should provide a reliable indication of the mean trajectory pattern under those conditions. Ultimately, statements about potential uncertainty/bias should be related to the spatial scales of ozone transport patterns inferred from the results.
Scientific Implications: Results of both location based and concentration based sorting of trajectories reveals the locations favored by trajectories associated with above average ozone concentrations at the trajectory end points (receptors). The central portion of the OTAG domain is such a favored location for receptors in many different locations throughout the eastern U.S. As noted above, however, these results do not indicate the relative contributions of different source regions to ozone levels at the receptor site.
Policy Implications: Emission controls in the areas suggested by the trajectory calculations to be consistently upwind of receptor sites experiencing high regional ozone levels may be effective in reducing regional ozone levels. For many of the receptor sites examined throughout the eastern U.S., the area implicated is centered on the Ohio River Valley. However, this conclusion cannot be confirmed solely on the basis of these studies given the limitations of the methodology discussed above.
Draxler, R.R. 1992: Hybrid Single-particle Lagrangian Integrated Trajectories (HY-SPLIT): Version 3.0 - User’s Guide and Model Description. NOAA Technical Memorandum ERL ARL-195, Air Resources Laboratory, Silver Spring Maryland.
Phillips, 1975: NOAA Technical Report NWS-22
Rolph, 1992: NGM Archive. NCDC Report TD-6140, National Climatic Data Center, July, 1992.
Schichtel, B. and P. Wishinski, 1996. HY-SPLIT and CAPITA Monte Carlo Model Back Trajectory Comparison. Summary Draft, 1 July 1996 (http://capita.wustl.edu/otag/Reports/Trajcomp/trajcomp.html)
Source Regions of Influence for High and Low Ozone Conditions in the Eastern US
Participants: Bret Schichtel and Rudolf Husar, Washington U., St. Louis, MO.
References: Update on the Characterization of Transport over the Eastern US. Schichtel and Husar, February 1, 1997.
Source Regions of Influence for High and Low Ozone conditions in the Eastern
U.S.,Schichtel and Husar, 8/08/96 Summary Draft.
Purpose: Develop a plume based method for investigating the spatial impact of ozone and ozone precursors from selected source areas (the source region of influence). Apply the technique to the Eastern US to characterize potential source regions of influence and transport patterns during the summers (June _ August) of 1991_95 for both average conditions and periods of high and low ozone concentrations, as well as during the 1991, 1993, and 1995 OTAG episodes.
Method: Forward particle plumes were calculated using the CAPITA Monte Carlo Model (Schichtel, 1995) with archived wind fields from the NGM weather forecasting model. The forward plumes represent simulations of air parcel transport away from a source, showing the direction and speed of the transport. Ranges of influence downwind ("source regions of influence") were estimated by imposing one or two day lifetimes on the released particles. By aggregating trajectories for multiple days the authors obtained climatological transport directions, air parcel residence times, and regions of influence. Naturally, the region of influence was larger for an assumed two day lifetime than a one day lifetime. Two methods were used for visualizing the results. In the first method, six sites were selected throughout the OTAG domain and the source region of influence around each site plotted. The region of influence is elongated along the predominant transport direction(s) and, viewed together, the six regions of influence convey a sense of the main transport directions and the one/two day regions of influence. In the second method, 504 sources were evenly distributed on a grid covering most of North America and one day regions of influence were calculated around each source. A vector was plotted at each source showing the characteristic (i.e., predominant) transport direction where the length of the vector was a measure of the persistence of transport in the direction of the vector. The resulting plot has the appearance of a gridded wind field plot. In addition, residence times were computed as the inverse of the characteristic transport speed. The characteristic transport speed was in turn estimated by dividing the radius of circle with area equal to the source region of influence by the assumed pollutant lifetime.
Findings: The source regions of influence are highly dependent upon the pollutant lifetime, and, as the lifetime of ozone is difficult to characterize, it is not possible to interpret the source regions of influence in a quantitative manner. The results can be used to qualitatively interpret relative changes in the size of regions of influence under different conditions. Results aggregated over all days (June_August in 1991-95) suggest two characteristic transport regimes over the Eastern US. One is from Texas north, through the Midwest and then east through New England to the Atlantic. The other regime is in the Southeast with shorter transport distances. These regimes are consistent with summer mean wind fields over the eastern U.S. which are dominated by the Bermuda High pressure system. Mean residence times increase approximately 50% from the northern half of the OTAG domain (~0.2 s/m) to the southern half (~0.3 s/m).
Aggregating results for days with the highest and lowest 10 percent average daily maximum ozone at each grid point showed that high ozone in the Southeast is associated with longer residence times indicating relatively stagnant air masses with net transport from the north. Low ozone is associated with shorter residence times and more persistent transport from the Gulf of Mexico and the Atlantic Ocean. In the Midwest, high ozone conditions are also associated with relatively long residence times indicating stagnant conditions under which a high pressure system is located over the Midwest. Low ozone in the Midwest is associated with shorter residence times and more persistent, rapid flow from the West. In contrast, high ozone episodes in New England are associated with persistent, relatively rapid transport (relatively short residence times) from the west-southwest whereas low ozone is associated with transport primarily from the northwest. Overall, residence times are lower than average during the low ozone events with transport from outside of the OTAG domain inward. While the high ozone events have longer than average residence times with transport from the center of the OTAG domain outward.
Calculations of residence times and transport vectors during each of the 1991, 1993, and 1995 OTAG episodes shows different spatial patterns with some similarities between the 1991 and 1995 episodes. The 1993 episode is characterized by long residence times in the south as was the case for the other two episodes but also by shorter than average residence times in much of the rest of the domain.
Limitations: The techniques developed are quite effective at illustrating climatological transport patterns, but less effective at showing regions of influence because it is unclear what lifetime to assume. The analyses do not shed light on the extent to which transport actually was important under different conditions, they can only address the potential for transport. The vector plots focus on the predominant transport direction, but the region of influence plots show significant potential for transport in different, even opposing directions. Therefore, as noted by the authors, the vector plots may obscure important details of the transport patterns. For instance, based on the study findings, a plausible scenario for significant transport from the Midwest to New England might be a stagnation event in the Midwest followed by the onset of organized transport from the Midwest to the New England, but it is unclear whether the analyses performed would be able to detect such an interaction. The potential effects of trajectory uncertainty and bias should be discussed in the context of the study conclusions.
Implications: This study suggests the potential for transport from the Gulf Coast to the Midwest and the Midwest to the Northeast, but does not provide information on whether such transport actually occurs. High ozone events in the southeast and Midwest appear to be associated with relatively stagnant conditions, which suggests that ozone levels in theses areas are less strongly influenced by inter-regional transport. However, the net transport under these conditions is from the Midwest suggesting potential contributions to high ozone in the South.
Ozone/Tracer/NOY Relationships at Three SOS-SCION Sites
Participants: Eric S. Edgerton and Benjamin E. Hartsell, ESE Environmental Inc. (Currently affiliated with Atmospheric Research and Analysis Inc.)
References: Slides from presentation given to the OTAG-AQA meeting, Washington D.C., July 1996 and written summary posted to the OTAG-AQA web site on November 4, 1996.
Objective: Identify the sources contributing to elevated ozone at SCION sites in the South by using CO and SO2 as tracers to distinguish between contributions from urban areas and major point sources. Investigate the spatial extent of elevated ozone events by looking at the coincidence of elevated ozone at different sites.
Methods: Data for the 1995 ozone season (may to October) from three sites in the SCION network in the rural south were analyzed. The sites were: Oak Grove (about 120 km NW of Mobile, AL), Centreville (about 85 km SW of Birmingham, AL) and Yorkville (about 65 km NW of Atlanta, GA). The measurements were for O3, NOY, CO and SO2 and were for 1 minute averaging times. Elevated ozone days were defined as those having 8-hour average ozone greater than 80 ppb. These days were classified as having high ozone due to urban emissions, major point source emissions, or mixed (i.e., periods of both urban and major point source impacts within the same day) by examining CO/NOY and SO2/NOY ratios. Periods with relatively high SO2 and low CO were attributed to major point sources, whereas periods with relatively low SO2 and high CO were attributed to urban emissions. An 80 ppb background for CO was subtracted before calculating the CO/NOY ratios.
Findings: All three sites observed elevated ozone days in the 1995 season, but there were fewer at Oak Grove (5) than Centreville (19) or Yorkville (29). Elevated ozone was observed at one or more site(s) on 39 days, of which 13 involved 2 sites and one involved all three sites. Therefore, about two thirds (25/39) of the elevated ozone days involved only a single site, suggesting that on these days the spatial scale of elevated ozone was less than the distance between any two sites (250-300 km).
A strong relationship between O3 and NOY was observed at all sites on clear sunny days. There was some scatter in the relationships and there were instances of departure from the predominant relationship. There was evidence of non-linearity in the relationship with a lower slope of O3/NOY at higher NOY (5-10 ppb) than lower NOY (1-5 ppb).
It was possible to distinguish periods of urban and major point source plume impacts at all three sites. At Centreville and Yorkville, there were elevated ozone days classified urban, major point and mixed; but at Oak Grove all five elevated ozone days were classified as mixed. Across the network the rank order of classifications was urban > mixed > major point.
Urban and point source plumes could be distinguished at these sites because key tracer species had high degrees of covariance. This suggests that the plumes were sufficiently recent in origin to retain their physical identity, i.e., they had not yet been dispersed into the regional airmass. The age of the plumes can’t be estimated directly from these data, however, the locations of the monitoring sites relative to nearby major urban areas suggests that the urban plumes must have been transported at least 65 to 120 km to influence these sites.
Limitations: The implications of the results are limited in the number of sites (3), the geographic distribution of sites (rural southeast) and the number of years investigated (1995). The unfortunate paucity of data required for this type of analysis prevents generalization to other parts of the OTAG domain. On the other hand, the data considered appear very rich mainly by virtue of the high temporal resolution (1-minute) which permit robust inter-species correlation analyses. The high degrees of covariance between species during elevated ozone periods may suggest that the plumes were of recent origin, but there is no way to estimate how recent or therefore what range of influence is implied.
The selection of elevated ozone days was based on an 8-hour period which may tend to focus away from days that had short term impacts. Point source plumes might be expected to result in shorter term impacts than urban plumes by virtue of there smaller (narrower) spatial scale, and this is born out by some of the data presented by the authors. Therefore, the basis for selecting elevated ozone days may influence the relative numbers of days classified as being urban, point or mixed influence.
Since the Oak Grove site is much closer to the Gulf Coast than the other two sites, it can be expected to have a quite different ozone climatology due to the influence of marine air. Therefore, looking at the simultaneous occurrence of elevated ozone at all three sites may not be a good way of examining the regionality of high ozone episodes in South. Looking for concurrent events at Centreville and Yorkville may provide a somewhat more reasonable estimate of frequency of periods of widespread high ozone in the south. Better still would be to use the more extensive multi-year database of routine ozone measurements.
Scientific Implications: The analyses performed in this study provide a good example of how tracer SO2/CO/NOY analyses can be used to distinguish point and urban source signatures in relation to ozone formation. Much of the power of the tracer analyses stems from the high temporal resolution of the SCION data. The data strongly suggest that both urban emissions and point sources can cause elevated ozone levels at these three sites. Furthermore, urban emissions played a more dominant role than major point source emissions for the kind of days selected (8 hour O3 > 80 ppb) at these three sites.
Policy Implications: For the three southern sites considered here, the majority of days with 8-hour ozone exceeding 80 ppb were associated with urban signatures, sometimes in combination with point source signatures. There were also some days when 8-hour ozone exceeded 80 ppb that had predominantly point source signatures. These findings suggest that controls on urban area emissions may be more effective at reducing the frequency of such exceedances than controls on point source emissions for this set of sites and this type of ozone episode.
Analysis of Ozone, NOY and Tracer Data from a Site in South-Central Pennsylvania
Participants: Eric S. Edgerton and Benjamin E. Hartsell, ESE Environmental Inc. (Currently affiliated with Atmospheric Research and Analysis Inc.)
References: Written summary posted to the OTAG-AQA web site on January 6, 1997
Objective: Identify types of sources contributing to elevated ozone at a rural site toward the West of the Ozone Transport Region (OTR) by using CO and SO2 as tracers to distinguish between contributions from urban areas and major point sources.
Methods: The Arendtsville site is located about 90 km Northwest of Baltimore in a rural area toward the West of the OTR. The site is located on top of a small hill and there is no major point source of SO2 or NOx is within 40 km of the site, and no secondary road within 200 m of the site. The data include continuous measurements of O3, NO, NOY, CO and SO2 reported at 1, 15 or 60 minute averaging times recorded between June 17 and September 30, 1995. Elevated ozone days were defined as those having 8-hour average ozone greater than 80 ppb. These days were classified as having high ozone due to urban emissions, major point source emissions, or mixed (i.e., periods of both urban and major point source impacts within the same day) by examining CO/NOY and SO2/NOY ratios. Periods with relatively high SO2 and low CO were attributed to major point sources, whereas periods with relatively low SO2 and high CO were attributed to urban emissions. An 80 ppb background for CO was subtracted before calculating the CO/NOY ratios. The cut-points for assigning episode types were:
Urban >10 <0.5
Mixed >5 >0.5
Major Pt. Source<5 >1.5
These cut-points were used to determine if one type of source was probably the dominant contributor, meaning that the source type could have provided at least 75 percent of the NOY. This analysis is similar to one performed by the same authors for three southern sites and described in the summary "Ozone/Tracer/NOY Relationships at Three SOS-SCION Sites."
Findings: Of the 101 days in the 1995 field season, 21 had 8-hour average ozone greater than 80 ppb. As in the analysis of data from the SOS-SCION sites, the authors found that ozone was correlated with NOY levels on high ozone days, and that there were correlations between NOY and CO and or SO2. The high temporal resolution in the concentration data greatly enhance the ability to discern such correlations. However, compared to the SOS-SCION sites, there were fewer periods that were clearly dominated by one type of source in the Arendtsville data. Most of the days were diagnosed as having "mixed" influence (i.e., both urban and major point source emissions contributed to the high ozone), a few (six) were diagnosed as urban dominated and none were diagnosed as major point source dominated. Thus, even for a site at the western edge of the OTR, urban emissions were associated with all 8-hour averages > 80 ppb.
Limitations: The implications of the results are limited because only one site in the Western OTR was evaluated. Caution should be used in extrapolating the results for Arendtsville to the Western OTR in general. The selection of "high ozone days" was based on an 8-hour period and a threshold of 80 ppb. An analysis for ozone events that are of shorter duration but higher concentration (e.g., 1 hour, 120 ppb) might reflect a different mix of sources.
Six of the 21 high ozone days were classified as likely urban dominated using a criterion of SO2/NOY < 0.5, but 5 of these 6 days had SO2/NOY ratios that were between 0.4 and 0.5. Thus, a relatively small change in this classification criterion to SO2/NOY < 0.4 would result in only 1 of 21 days being classified as urban with the remainder all classified as mixed. Similarly, several of the "mixed" days exhibited SO2/NOy ratios of about 0.6 and CO/NOy ratios greater than 10 such that, if the criterion for SO2/NOy ratios were a little higher, there would be a few more days classified as urban.
It appears that high ozone days at Arendtsville usually are influenced by a mix of urban and point sources. The authors investigated several approaches to estimating the relative contributions of these source types by analyzing ratios of pollutant concentrations, but the analyses are complicated by the need to account for the effects of atmospheric removal on the CO, SO2 and NOY concentrations measured at Arendtsville. These analyses contain significant uncertainties because the age of emissions impacting Arendtsville, and the atmospheric lifetimes of NOY, SO2 and, to a lesser extent, CO, in the surrounding environment are hard to quantify.
Scientific Implications: The analyses performed in this study provide a good example of how tracer SO2/CO/NOY analyses can be used to investigate the contributions of point and urban sources to ozone formation at a site where high temporal resolution data are available. Data for Arendtsville, in the western OTR, indicate that a mix of urban and point source emissions causes elevated ozone levels. Urban emissions appear to be involved in all episodes, while point source emissions appear to be involved in the majority of episodes. The O3 to NOY ratio measured at Arendtsville was about 4, which suggests that ozone was formed under conditions that were VOC limited or intermediate between VOC and NOX limited. Ratios at Arendtsville should be compared to other sites in the western OTR to investigate whether the relatively low values measured are representative of air masses transported into the OTR or whether local emissions may have influenced the Arendtsville site.
Policy Implications: For Arendtsville, almost all days with 8-hour ozone exceeding 80 ppb were associated with mixed urban and point source signatures, manifested by characteristic ratios of CO and SO2 to NOy. This suggests that controls on urban or point source emissions have the potential to reduce elevated ozone levels at Arendtsville. Given the rural nature of this site, as well as its location in the western OTR, it is reasonable to expect that the urban character of episodes would be more dominant in most other parts of the OTR.
A Comparison of Modeled and Measured Ozone, NOY and CO at Nine Regional Monitoring Stations during the 1995 OTAG Episode
Participants: Benjamin E. Hartsell and Eric S. Edgerton, ESE Environmental Inc. (Currently affiliated with Atmospheric Research and Analysis Inc.)
References: Written summary posted to the OTAG-AQA web-site on November 24, 1996.
Objective: Use data from 9 rural monitoring stations to evaluate the OTAG UAM-V model performance for the 1995 episode. The questions addressed are how does the model chemistry perform (in terms of O3/NOY) and how well does the model predict observed O3, NOY and CO levels. Additional analyses currently in progress will include additional measurements and look at some issues in more detail.
Methods: Air quality data were assembled for the 1995 episode (July 7-18) from four southern sites in the SOS/SCION network (sites in MS, AL, GA and TN) and five northern sites from the NARSTO-NE study (sites in PA, NY and MA). All stations measured O3 and NOY and four measured CO. All sites are described as being located in "large clearings with excellent fetch and were well removed from the influence of local sources." Model predicted concentrations for the site locations were from the OTAG D2 base case. 1-hour averages of observed and modeled data were compared.
Findings: Scatter plots of O3 against NOY across all sites and all hours between 1100 and 1900 LST were prepared for the observations and model predictions. The plots showed significant variation in the relationship for both the observations and model predictions, however the general appearance of the plots was similar. When the data were sorted by ozone concentration and put into ten bins containing equal numbers of points, linear relationships between O3 and NOY emerged for both the observations and the model. Within each bin, the distribution of NOY values tails toward high values and so this analysis tends to emphasize the lower NOY values that likely represent cases where ozone is most strongly NOX limited. Therefore, this analysis effectively compares the modeled and observed O3/NOY relationships (i.e., the ozone yield per NOX) under NOX limited conditions, finding good generally agreement between the model (slope 8.7) and observations (slope 9.2). There was more scatter in the modeled O3/NOY relationship than in the observed relationship, which is consistent with the model over predicting the frequency of fresh NOX emission impacts at these sites.
Plots were prepared of bias (modeled - measured) against modeled values with points averaged into 10 bins of equal numbers of points. These plots are difficult to interpret without scatter plots of the underlying data. For instance, the plots of ozone residuals against modeled ozone and ozone residuals against measured ozone are quite different: one plot suggests model over predictions of about 20 ppb for the highest bin while the other plot suggests model under predictions of about 5 ppb for the highest bin. A clearer picture is obtained from analyses of ozone values exceeding thresholds of 80, 100 and 120 ppb. For these rural sites, the modeled and observed ozone exceeded 120 ppb in only a few instances and did not exceed 100 ppb at all sites. This shows that ozone levels above 100 ppb represent the upper tail of the distribution of ozone values for these sites. The ratio of modeled to observed hours above 80 and 100 ppb was 342/259 » 1.3 and 102/59 » 1.7, respectively. The ratio of modeled to observed days with values exceeding 80 and 100 ppb was 47/39 » 1.2 and 19/11 » 1.7, respectively. This shows a tendency for the model to over-estimate the frequency and duration of high ozone values at these sites.
The time series for CO showed a tendency for over prediction of CO concentrations for two southern sites in GA and TN. CO levels were generally predicted better for the southern site in MS (close to the Gulf) and for the one northern site in PA. The time series for NOY showed generally good agreement with some tendency for the model to predict higher maximum NOY values than were observed for the northern sites.
Limitations: The paper acknowledges several limitations that apply generally to model performance evaluations of this type: (1) observations at single points are compared to model predictions that are volume averages over large grid cells. Grid cells were about 12 km by 12 km by 50 m at most sites, but 36 km by 36 km by 100 m for the MS site. (2) The number of sites and days available for comparison is limited. Relative to other analyses of model performance for OTAG episodes, this limitation is more severe for ozone than the other species because there are a large number (hundreds) of other ozone sites available but only a small number of NOY and CO sites are available.
The modeled NOY does not include organic nitrates (Carbon Bond species NTR) because NTR concentrations were not saved in the OTAG model runs. Thus, modeled NOY concentrations are always biased low. However, the magnitude of this bias is not well known and likely varies in time and space depending upon the age of NOY (the bias might be as high as 20 percent for aged air masses). Accounting for the missing contribution of NTR to modeled NOY would increase the model under estimation of the O3/NOY relationship (perhaps by up to 20 percent).
The statistical analyses averaged data across all sites so that there is no indication whether significant differences existed for individual sites or between groups of sites (e.g., North vs. South).
Scientific Implications: The comparison of modeled and observed O3/NOY relationships revealed good agreement between the model (slope 8.7) and observations (slope 9.2). These values can be interpreted as estimates of the ozone yield per NOX under NOX limited conditions. However, the model value is biased high by an uncertain amount because organic nitrates were not included in the model NOY values: If possible, the magnitude of this bias should be investigated to determine whether it significantly influences this finding. Also, the comparison of modeled to measured O3/NOY relationships suggests that the model may be over predicting the frequency with which fresh NOX emissions impact these sites.
For these nine rural sites, there was a tendency for the model to over-estimate the frequency and duration of high ozone values: this should be investigated for other rural sites to find whether it is indicative of a more general model performance problem. The apparent tendency for the model to over predict CO at two sites in GA and TN may be indicative of emissions or atmospheric mixing problems in this region, and should be investigated.
Policy Implications: Analyses for nine rural sites in the South and the Northeast indicate that the model has a tendency to over predict the frequency and duration of ozone levels in 80-100 ppb range. If confirmed by more extensive analyses, this finding would indicate a potential problem with using OTAG results to provide boundary conditions for urban scale modeling. The good performance of model chemistry for O3/NOY relationships under NOX limited conditions suggests that overall there is no gross problem with the model chemistry.
Ambient Monitoring Sites for OTAG (time series) Model Evaluation
Participants: Rich Poirot, Vermont Department of Environmental Conservation. Vermont Air Program, Building 3 South, 103 S Main Street, Waterbury, VT 05676
Reference: Ambient Monitoring Sites for OTAG (time series) Model Evaluation, 11 January 1996 memo from Rich Poirot, Vermont Department of Environmental Conservation to S.T. Rao and Rudy Husar Available via WWW at "http://capita.wustl.edu/OTAGActivities/AQADocuments/OTAGdiscdocs/ModelCompSites.html".
Purpose: Recommend ozone monitoring sites in the Northeast to be used for evaluation of the OTAG base case regional photochemical model simulations for the July 1988, 1991, 1993, and 1995 episodes.
Methodology: Four ideal objectives were identified for site selection: (1) Sites should be spatially representative of approximately 12 x 12 km2 model grid squares; (2) Data should be available for all OTAG model episodes; (3) Data should include hourly measurements of ozone, NOx, and speciated hydrocarbons and carbonyls; and (4) Good spatial coverage of the modeling domain. The author noted that these four criteria for site selection are to some degree in conflict with each other. For example, the PAMS data sites have more complete chemical species measured but are more urban oriented (i.e., aligned along the Northeast Corridor) and primarily have data available only for the 1995 episode although some data are available for the 1993 episode as well.
Findings: Given the conflicting objectives noted above, the author developed a list of 21 ozone sites in the OTR which meet objectives 1, 2, and 4 (with some NOx measurements) and an additional set of 13 PAMS/NARSTO sites where more complete chemical species data are available but which primarily operated only during the 1995 episode (with some data possibly available at Whiteface Mtn, NY and Wye River and Shenandoah NP, VA for 1993). The author suggests that if this list is too long, priority should be given to: a)sites above 500 meters; b)sites with NOx data; and c)rural PAMS sites.
Limitations: Although the site selection approach used in this study provides a means of selecting optimal sites for evaluating the rural/regional model predictions, the emphasis on rural sites with broad spatial representativeness (e.g., sites above 500 m or rural PAMS sites) tends to preclude recommendation of sites in nonattainment areas. As the fundamental purpose of OTAG is to identify options for reducing the transport of ozone and precursors into nonattainment areas, OTAG must also be concerned with model performance on the edges of and within nonattainment areas since poor performance in these areas would throw into question the accuracy of model transport estimates.
Scientific Implications: Statistics summarizing model performance are likely to be highly dependent on the choice of sites to include in the analysis. The author is able to identify a number of predominantly rural sites that meet his criteria. However, as noted above, OTAG must also be concerned with model performance in and near nonattainment areas.
Policy Implications: Meeting OTAG’s goals requires development of photochemical model simulations that meet acceptable performance criteria. Appropriate selection of data sets to be compared with model predictions is critical to the calculation of model performance statistics relevant to OTAG’s goals. This study has identified monitoring sites in the northeastern portion of the OTAG domain that can be used to summarize model performance in rural and remote areas. Selection of sites in nonattainment areas and in other portions of the OTAG domain are not addressed here.
Comparison of OTAG UAM-V/BEIS2 Modeling Results with Ambient Isoprene Observations
Participants: Eric Edgerton and Ben Hartsell, Atmospheric Research & Analysis, Inc.; Ralph Morris, Keane Lee, and Greg Yarwood, ENVIRON International Corporation.
References: Comparison of Modeled versus Observed Isoprene Concentrations at Rural and Suburban Sites Across the Eastern U.S. E. Edgerton, 1997 (http://capita.wustl.edu/OTAG/Reports/Isoprene/ISOPRENE.html)
Phase I Comparison of OTAG UAM-V/BEIS2 Modeling Results with Ambient Isoprene and Other Related Species Concentrations. R. Morris, K. Lee, and G. Yarwood, 31 January 1997 (http://capita.wustl.edu/OTAG/Reports/morris/EXECSUM.html).
Purpose: Estimates of biogenic emissions used in the OTAG UAM-V modeling are based on the BEIS2 emissions model. BEIS2 isoprene emissions are significantly higher than those generated by its predecessor model, BEIS1, resulting in a biogenic emissions estimate (of which isoprene is the largest single component) that constitutes 70 - 80 percent of the total VOC emissions in the OTAG modeling domain. Thus, control strategy impact estimates obtained from the OTAG modeling depend critically on the accuracy of the BEIS2 biogenic emission estimates. Two independent analyses of the available ambient data were conducted to evaluate the UAM-V/BEIS2 isoprene concentration predictions.
Method: UAM-V/BEIS2 predictions of ambient isoprene concentrations in surface layer grid cells (roughly 12 km by 12 km horizontally by 50 m in the vertical) corresponding to ambient VOC monitor locations were compared with observed afternoon isoprene concentrations for the July, 1995 OTAG episode. Isoprene observations were obtained from ambient speciated hydrocarbon measurements made at PAMS network monitoring sites as well as sites operated as part of the NARSTO-Northeast and SOS-Nashville Intensive field studies. These sites are primarily concentrated in the Northeast and in the Tennessee-Kentucky areas where the two field studies were conducted in 1995. Additional sites are located in Virginia (one), Georgia (one), Baton Rouge, LA (two), and western Pennsylvania (one). Other portions of the OTAG domain (notably the Midwest, Mid-Atlantic region, Gulf Coast outside of LA, southern Florida, and Texas) are not well represented by these databases. Edgerton’s analysis concentrated primarily on 16 rural and suburban sites while Morris et al. focussed on 15 urban/suburban (or urban influenced) sites. The two studies have six sites in common. At some sites, (primarily PAMS and NARSTO-NE) observations were made every hour by automated gas chromatographs (auto-GCs), while at others, integrated canister samples were collected at set times. For the auto-GC data, comparisons with model predictions were based on 11:00 - 17:00 LST averages (11:00 - 19:00 LST in the Morris et al. study); at the canister sites, afternoon integrated canister samples were compared with model predictions for the corresponding time intervals. In this way, comparisons were restricted to the period of maximum isoprene emissions and turbulent diffusion when comparisons of surface-based observations (5 - 15 m agl) with grid cell (volume average) predictions are most meaningful. All comparisons were done on a site-by-site (spatially matched) basis.
Table 1. Summary of bias and gross error results.
Morris et al.
2 - 21 ppbC
1.1 - 34.9 ppbC
Distributions of the biases across sites are summarized in Table 2.
Table 2. Distribution of bias across sites.
Under prediction greater than 50%
Over prediction greater than 50%
No. of Sites
% of Sites
No. of Sites
% of Sites
No. of Sites
% of Sites
No. of Sites
% of Sites
Morris et al.
Results: Afternoon observations included in the analysis showed a wide range of concentrations from near zero to over 60 ppbC; mean values across the network ranged from 2 ppbC to almost 30 ppbC. Comparison with predictions matched in time showed a wide range of prediction errors. Bias and gross error results are summarized in Table 1. Results are quite similar for both studies with maximum positive biases (over predictions) being larger in an absolute sense than the maximum negative biases (under predictions).
Bias was within ±50% at just 6 of the 16 sites examined by Edgerton, with all but one (Brookhaven, NY) of the remaining sites exhibiting over predictions. While bias was within ±25% at just two sites in each study, slightly better agreement overall was found by Morris et al.: just over half of the 15 sites examined showed biases within ±50 percent. One site (Ware, MA) under predicted by more than 50 percent, with over predictions of greater than 50% at the remaining six. Thus, more than three-fourths of the sites examined in these studies exhibited biases greater than ±25% and nearly half the sites exhibited mean biases of greater than ±50% of mean observed values. Nearly all the biases over ±50% were over predictions. Edgerton found no discernable geographic pattern of bias. Edgerton’s analysis of residuals (mean differences as a function of mean predicted) indicates a fairly strong linear relationship (r2 of 0.78) with a positive slope: of the eight sites with mean predicted concentrations below 10 ppbC, the model under predicted on average at three sites while for the eight sites with mean predictions above 10 ppbC, all but one exhibited over predictions on average. While this relationship is partially a reflection of the lack of correlation between observed and predicted isoprene concentrations, the strength of the correlation between the differences and predicted values in this case suggests that some additional factor is contributing to the relationship. Thus, in an absolute sense, model performance is worse at sites with high mean isoprene concentrations (e.g., in the southeast) than at sites with low mean concentrations. Morris et al. did not conduct a similar analysis of residuals.
Edgerton examined two potential sources of bias between observations made at a fixed point roughly 10 m above the surface and model predictions made for a 50 m deep grid cell: 1) the decrease in isoprene concentrations with height (which would cause predicted values to be less than observed), and 2) the use of temperature measurements made at roughly 1.5 m agl to estimate isoprene emissions from forest canopies located further above the ground where cooler temperatures can be expected as a result of near adiabatic or super adiabatic lapse rates typical of sunny summer afternoons (which would cause predicted isoprene values to be greater than observed). In some cases, forest canopy temperatures may also be lower than the near-surface observations due to enhanced evapotranspiration. Edgerton’s review of the isoprene profile estimates of Andronache and Chameides (1994) indicates that observations taken at 10 m should be only about 2 - 6 percent higher than the 10 - 50 m layer average. As for the temperature effect, Edgerton states that daytime vertical temperature gradients in rural eastern U.S. environments as measured at CASTNet meteorological monitoring sites are typically -0.5 to -1.0 deg. C between 2 m and 9 m. If this is assumed to be the same order of magnitude as the temperature difference between 1.5 m and the vegetation canopy, it could cause a possible overestimation of emissions on the order of 7 to 15 percent. Thus, the influence of vertical gradients in isoprene and temperature appear to be negligible in comparison to the relatively large biases between observed and predicted isoprene concentrations found in these studies. Of course, this does not rule out the possibility of other temperature-related problems such as biases in the 1.5 m measurements used to generate gridded wind fields for the model.
Limitations: Comparison of observed and predicted isoprene concentrations are primarily limited by the spatial representativeness of the monitoring data as well as the ability of the model to characterize vegetation within a grid cell. Isoprene emissions exhibit strong spatial and temporal variations due to differences in the amount of plant material and species mix and the response of plants to changes in environmental conditions. In addition, isoprene is highly reactive and undergoes rapid transformation once released. Thus, measurements made at fixed locations with varying degrees of exposure to isoprene sources can provide only an approximate indication of the volume average concentration represented by the model predictions. Additional analysis of both the ambient measurements and the land use and temperature data input to the model are needed to better diagnose the factors contributing towards the prediction errors noted in these two studies and to rectify the apparent inconsistencies between these results and the emission flux measurements used to support development of BEIS2 (see Guenther et al., 1996 and references therein). If the underlying formulation of BEIS2 is sound as suggested by the earlier studies, then the OTAG over predictions point to a potential problem with the procedures used to calculate the OTAG biogenic inventory. It should also be noted that isoprene represents only one component of the total biogenic VOC inventory, all though it is the dominant species. Similar comparisons for other biogenic species are not included in these studies (and may be difficult or impossible to do for some species that are not unique to biogenic sources or are difficult to measure accurately).
Scientific Implications: Because isoprene is so abundant in the OTAG domain and as a result of its high reactivity, significant errors in isoprene predictions have the potential to introduce significant errors into predictions of the sensitivity of ozone to anthropogenic VOC and NOx controls. Isoprene is a highly reactive primary species with emissions that vary strongly as a function of solar insolation, temperature, and land cover. As a result, one might expect relatively large differences between volume average predictions made for a 12 by 12 km by 50 m deep grid cell and measurements made at a single fixed location within that cell. Measurement errors would also contribute to these differences. Results of these studies bear out this expectation, showing much larger differences between UAM-V/BEIS2 isoprene predictions and corresponding observations than is the case for ozone, even when the comparison is limited to the afternoon hours when high emission rates and strong turbulent diffusion should act to minimize the expected differences. This expectation not withstanding, the magnitude of the prediction errors and the tendency towards over predictions indicates a potentially serious problem with the UAM-V/BEIS2 modeling system. The fact that over predictions occur at many of the rural sites examined by Edgerton is especially troublesome - over predictions at some urban sites could be explained by the fact that the corresponding modeling grid cells include vegetation cover beyond the urban boundaries. Isoprene emission rate biases for tall forest canopies introduced by the use of inappropriately high near surface temperatures in BEIS2 do not appear to be large enough to explain the observed over predictions, although other potential sources of error in the temperatures used to drive the model have yet to be explored.
Policy Implications: Incorrect representation of biogenic VOC emissions in the OTAG UAM-V/BEIS2 modeling system would result in errors in the model’s predictions of ozone concentrations and the model’s responses to anthropogenic emission control strategies. If isoprene emissions in the OTAG inventory are significantly overestimated (for whatever reason), the over representation of biogenic VOCs would cause the model to underestimate the expected response to anthropogenic VOC controls. In addition, the model would tend to be overly sensitive to NOx controls in all but the largest urban areas.
Andronache, C. and W.L. Chameides, 1994. Vertical distribution of isoprene in the lower boundary layer of the rural and urban southern United States. J. Geophys. Res. 99:16989-16999.
Guenther, A., P. Zimmerman, L. Klinger, J. Greenberg, C. Ennis, K. Davis, W. Pollack, H. Westberg, G. Allwine, and C. Geron, 1996. Estimates of regional natural organic compound fluxes from enclosure and ambient measurements. J. Geophys. Res. Vol. 101, p. 1345.
Comparison of SOS Nashville Data to OTAG 1995 Base Model
Participants: Robert E. Imhoff, Tennessee Valley Authority
References: Summary of, and slides from, presentation given to the OTAG AQAWG meeting at Cherry Hill, NJ, August, 1996.
Purpose: Use ambient data from the 1995 Nashville SOS study to evaluate how well the OTAG air quality modeling for July 9-18, 1995 is simulating atmospheric chemistry and physics in a southern environment.
Methodology: UAM-V results for the July 9-18, 1995 OTAG D2 base case were obtained from the New York State Department of Environmental Quality. Model results were compared to three ambient data sets. The first data set was morning surface CO, NOY, and VOC concentration data for an urban site in Nashville for June 19 to July 28, 1995. The measured species ratios were compared to their relative proportions in the emissions inventory. The second data set was for ozone, NOX and NOY at a suburban and a rural site. Ozone productivity (i.e., the relationship of O3 to NOZ where NOZ = NOY - NOX and where O3/NOZ is the number of ozone molecules produced per NOX molecule lost) was assessed at these two sites by comparing measurements and model results during the 11:00 to 18:00 daytime period when the atmosphere was well mixed. The third data set was for vertical profiles of ozone over a suburban site near Nashville. The accuracy of vertical mixing processes in the model was assessed by comparing measured and modeled ozone profiles. The ozone sonde data were collected by Georgia Institute of Technology, the remainder of the data were collected by TVA.
Findings: For morning surface concentration data at the Nashville urban site, measured CO/NOY slopes were 7.7 for June 19 to July 28 and 10.6 for July 9-18. The modeled slope for July 9-18 was 12.6, and the emission inventory slopes were 8.2 for mobile sources and 7.5 all sources. The apparent background of CO in the ambient data was 150 - 200 ppb, compared to about 100 ppb in the model. The measured VOC/NOY slope for July 9-18 was 2.5 compared to a modeled slope of 2.9. The apparent background of VOC in the ambient data was quite close to zero, compared to about 90 ppb in the model. Agreement between the modeled and ambient CO/NOY and VOC/NOY slopes was reasonably good. Modeled CO/NOY slopes would be expected to be higher than inventory slopes because VOC oxidation is a secondary source of CO, and because NOY is deposited more rapidly than CO.
Comparing observed and modeled ozone concentrations at the suburban and rural sites revealed a fair amount of variability between the model and observed ozone on a day to day basis, but no indication of bias overall. At the rural site the model estimated the ozone productivity very well, with least squares regressions for both model and observations showing O3/NOZ slopes of about 6, consistent with NOX limited ozone formation. At the suburban site, there was more variability in both observed and, in particular, modeled O3/NOZ ratios. The observed ozone productivity (least squares regression slope about 3.5) was lower than at the rural site consistent with ozone formation being less NOX limited at the suburban site than at the rural site. The model ozone productivity at the suburban site was about 4.3 on July 9-17 which is similar to the observed ozone productivity (about 3.5). However, on July 18 the model O3 concentrations were much lower than on July 9-17 resulting in lower O3/NOZ ratios.
Further investigation of the model performance for the suburban site (Youth) on July 18 showed that the model predicted high concentrations of NOX (10-15 ppb) between 11:00 and 18:00, contrary to the observations. In other words, the model predicted that Youth was impacted during the middle of the day by a substantial source of fresh NOX that was not observed on July 18, or on July 9-17 for that matter. As a result of this error, the model predicts ozone formation at the Youth site to be VOC limited on July 18, whereas the observations show the site to be more NOX limited. Thus the model sensitivity to VOC versus NOX emission control around the Youth site would likely be opposite to that predicted by the observations for this day.
The comparison of observed and modeled vertical profiles for ozone showed that the model consistently overestimated the surface and layer 4 ozone concentrations (measured layers 2 and 3 were not available for comparison because the resolution of the sonde was too coarse in these layers). The model consistently estimated a slight increase in ozone from layer 1 through layer 4, with a sharp drop above layer 5. The top of the model's mixed layer was in layer 5 during this period. The model's estimated ozone concentration in layers 6 and 7 were consistently too low. One possible cause is that the previous day's mixed layer in the model was too low and thus material was not mixed to layers 6 and 7. Another possibility is that the model does not handle localized mixing due to convective cells which were common during the meteorological conditions prevalent during this period. These cells can inject significant amounts of air from near the surface above the mixed layer. Once above the mixed layer the material can have long residence times. Another possibility is that the "Topcon" concentration used at the upper boundary in the model (35 ppb) is too low.
Limitations: The modeled NOY used in these comparisons does not include organic nitrates (Carbon Bond species NTR) because NTR concentrations were not saved in the OTAG model runs. Thus, UAM-V NOY concentrations are always biased low. However, the magnitude of bias is not well known and likely varies in time and space depending upon age of NOY (the bias might be as high as 20 percent for aged airmasses). Accounting for the missing contribution of NTR to modeled NOY might: a) further improve agreement between the observed and modeled CO/NOY and VOC/NOY ratios for the urban site; b) further improve agreement between observed and modeled ozone productivity at the suburban site, but; c) degrade agreement between observed and modeled ozone productivity at the rural site.
Scientific Implications: The relatively good agreement between modeled and ambient CO/NOY and VOC/NOY ratios supports the validity of these parameters in the emission inventory, in overall terms. However, the discrepancy between the apparent background for VOC of 90 ppb in the model compared to near zero in the ambient data possibly indicates a source of VOC in the model that is not also a source of CO. Further investigation is needed to identify the source of this discrepancy. The modeled and observed ozone productivities were consistent at the rural site, and were consistent at the suburban site on all but one day. The generally good agreement in ozone productivity suggests that the model chemistry is performing well for rural and suburban sites in the Nashville area. However, the inability of the model to correctly predict the ozone productivity at the suburban site on July 18 cautions that OTAG model performance is subject to important day-to-day variabilities that have the potential to influence emission control decisions. The discrepancy in the vertical ozone profile data may suggest that the model is not properly distributing mass vertically or that the top boundary concentration for ozone is too low. Underestimating vertical mixing might tend to overestimate concentrations at the surface, even though the chemistry and emissions inventory are accurate. Underestimating the top boundary concentration for ozone might have a small impact of biasing surface ozone concentrations low. While no prediction bias for surface ozone was noted at the sites examined in this study, it is interesting to note that, overall, the model was found to over predict surface ozone in the southeastern U.S. (EPA, 1996).
Policy Implications: Although not conclusive, this study raises the possibility that the UAM-V model runs are not treating concentrations in the upper layers of the modeling domain properly, thereby introducing errors in the estimated impact of transport on downwind areas. The inability of the model to correctly predict the influence of NOX emissions on ozone formation at the suburban site on one day sends a cautionary message for the use of OTAG model results at suburban, and presumably urban sites. Further technical analyses will be needed to sort through these issues.