Mar 13, 2012 - To cite this article: Douglas G. Smith & Bruce A. Egan (1979) Design of Monitor Networks to. Meet Multiple Criteria, Journal of the Air Pollution ...
Journal of the Air Pollution Control Association
ISSN: 0002-2470 (Print) (Online) Journal homepage: http://www.tandfonline.com/loi/uawm16
Design of Monitor Networks to Meet Multiple Criteria Douglas G. Smith & Bruce A. Egan To cite this article: Douglas G. Smith & Bruce A. Egan (1979) Design of Monitor Networks to Meet Multiple Criteria, Journal of the Air Pollution Control Association, 29:7, 710-714, DOI: 10.1080/00022470.1979.10470850 To link to this article: http://dx.doi.org/10.1080/00022470.1979.10470850
Published online: 13 Mar 2012.
Submit your article to this journal
Article views: 41
View related articles
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=uawm16 Download by: [190.42.216.132]
Date: 30 January 2016, At: 11:37
Design of Monitor Networks to Meet Multiple Criteria
Douglas G. Smith and Bruce A. Egan
Downloaded by [190.42.216.132] at 11:37 30 January 2016
Environmental Research & Technology, Inc.
For most monitoring networks, demonstration of compliance with ambient air quality standards is only one of many interrelated purposes served. The network also may be needed (1) to assess current air quality, (2) to assess regional background air quality, (3) to determine individual source "culpability," (4) to validate or calibrate a particular dispersion model, (5) to assess effectiveness of a control strategy, or (6) to determine the risk of damage to certain critical or sensitive receptors. Monitoring to support compliance with Prevention of Significant Deterioration regulations is now also an important issue for many new sources. The same network may have to perform many of these functions for several pollutants that have different source locations and different characteristic averaging times. This paper presents a method for systematically addressing each of these concerns. The method includes use of a statistical computer model, MONITOR. This model provides quantitative estimates of relative probabilities in order to assess whether a particular network design will meet each of these criteria. A sample application of the method is provided.
Air quality monitoring networks have been designed and operated for a wide variety of purposes. Major reasons have included: (1) assessing current (often preconstruction) local air quality; (2) identifying regional background air quality; (3) determining individual source "culpability"; (4) validating or calibrating dispersion models; (5) assessing control strategy effectiveness; and (6) determining risk of damage to especially sensitive receptors. A particular monitor network design strategy may, and for cost efficiency should, incorporate several of these purposes. For many new and proposed sources the new Prevention of Significant Deterioration (PSD) regulations are of principal importance for defining monitoring requirements. For those sources planning to locate or expand in an area currently attaining National Ambient Air Quality Standards (NAAQS), a regulatory requirement for preconstruction monitoring will only follow if the new source threatens to make the area nonattaining. "Screening" models are often used in making this assessment. It must be noted, however, that sources not required to conduct monitoring programs may still benefit from monitoring. The potential reasons still include several of those Copyright 1979-Air Pollution Control Association
710
outlined above, but with new emphasis on: (1) determining present compliance (particularly for areas designated as nonattaining); or (2) tracking increment consumption. This paper discusses the quantitative modeling methods for network design of effective single and multiple purpose monitoring systems. Monitoring Requirements for PSD Permits
All major facilities, defined by EPA guidelines1 as any of 28 source categories having a potential to emit at least 100 tons per year (tpy) of any pollutant regulated by the Clean Air Act, must at least consider the need for monitoring. (Other types of sources emitting greater than 250 tpy of these pollutants are also classified as major facilities.) If modeling or previous monitoring shows that an air quality impact analysis is needed according to PSD regulations,2 then ambient monitoring of the governing pollutants is required before a PSD permit can be obtained. For PSD applications, the most important function of a network is to define current air quality. The concentrations monitored define whether the site or the area of future plume impact are currently attaining the NAAQS. This status defines whether the NAAQS or the allowable PSD increments govern the source and the required methods for documenting expected impacts. It may not be surprising that the required preconstruction measurements may themselves change the definition of the attainment status of the area. For example, an area classified as a nonattainment region may be shown to be attaining in all but a small subarea that is not influenced by the proposed source. On the other hand, it is likely that many previously unrecognized nonattainment areas are likely to be discovered. The EPA guidelines for PSD monitoring are quite specific in discussing monitoring plans, as well as technical siting requirements (such as horizontal and vertical probe placement, or spacing from obstructions or local sources). But guidance as to the location and number of sites and pollutants monitored must remain general due to the site-specific nature of these decisions. The following sections describe the quantitative methods available for designing networks for PSD monitoring as well as the other rationales outlined above. Quantitative Methods for Network Design
Three types of quantitative methods used in siting monitors are discussed here. The first is the MONITOR computer model, which has been previously described elsewhere.3 It is based on a gaussian climatological model similar to, but Journal of the Air Pollution Control Association
simpler than that described by Miller and Noll4. MONITOR is used principally for assessing short-term concentrations from large point sources. The second is a sequential modeling method. It includes statistical analysis of 1 to 24 hr average concentrations and evaluation of annual average concentrations calculated for multiple point sources and/or area sources. The third method is described by Hougland and Stephens5 as a geographical coverage model. With some modification it can be thought of as a generalization of the other two methods, although it is similar to the first climatological model in its logical structure.
Area bounded by resolution
Area bounded by threshold
Downloaded by [190.42.216.132] at 11:37 30 January 2016
MONITOR—A Climatological Modeling Method
The monitor siting method uses a site-ranking computer program as one of several equally important analysis phases. The first step is to review the diffusion meteorology and climatology of the air quality region of interest. The second step is to decide whether the most readily available meteorological data are likely to be representative of the flows to be modeled, and whether a gaussian model is adequate. (For a coastal site, a sea breeze circulation may have to be modeled separately and may require use of meteorological data from additional measurement stations.) Climatological data are used to develop a matrix of frequencies of meteorological conditions, or classes, necessary for MONITOR analysis. Currently the MONITOR program is designed to accommodate 2400 classes of weather conditions (16 wind directions, 6 wind speed classes, 5 atmospheric stabilities, and 5 mixing depth classes). When daily mixing depth data are not available, a single mixing depth corresponding to a climatological mean estimate may be supplied, reducing the number of independent weather classes to 480. The third step is to use the selected dispersion programs to calculate locations of maximum ambient concentrations. With frequencies of meteorological conditions and resultant concentrations profiles as input data, the last step is to execute the MONITOR site ranking program and select the final sites for a network. The choice of dispersion model algorithms used to calculate ambient concentrations of pollutants is not critical to the MONITOR method, as long as the meteorological class matrix for which concentrations are calculated has exactly the same format conventions as the climatological frequencies. The example presented below employs a gaussian model with modeling assumptions similar to those in the EPA-RAM program, though only hourly average concentrations are calculated here. The Environmental Research & Technology, Inc. (ERT) Point Source Diffusion Model (PSDM) frequently used with MONITOR is a modular gaussian model which calculates concentrations for 2400 classes of weather conditions. Use of this model has the advantages, with respect to PSD analysis applications, of separate subroutines that can be used to (1) model background contributions to air quality, (2) calculate background based on data from existing monitors, and (3) evaluate contributions from point sources. Any or all of these contributions can be summed, so that the total concentrations profiles associated with each of the weather classes can be used as input to the MONITOR site selection program. PSDM can incorporate special subroutines to correct model calculations for flow over and around elevated terrain, and to simulate the trapping of plumes under conditions of limited mixing where the effective mixing depth varies throughout the day. To use the MONITOR program for ranking sites, one must supply six selection constraints. They are: • concentration threshold, • resolution of peak concentrations, • number of monitors to be ranked, • specific weather classes to be included, July 1979
Volume 29, No. 7
Concentration Peak - Resolution
~1
1
/ /
1
\
1 h
>>.
Threshold
1
Xs.'-Xi 1 . Coverage Ratio = — • L_ w hen Q = Q X2-X1 Figure 1a. Definition of area bounded by resolution and threshold.
•••X
X2
X'|
X|
X2
Downwind Distance Figure 1b. Frequency distribution of downwind distance intervals.
I
i
I I I
I
I I I
I Figure 1c. tion.
I I
^
Downwind Distance Cumulative frequency histogram for range selec-
711
data; the ranking of monitor locations by the same type of sequential model likely to be used in 24 hr standard compliance analysis of future measurement data yields a degree of consistency not otherwise possible. Coverage Models
Downloaded by [190.42.216.132] at 11:37 30 January 2016
• •
ranking criterion: frequency of exposure, and frequency of measurements required to justify station cost. A single location cannot be found, even within one wind direction sector, at which all occurrences of high concentrations can be monitored. MONITOR therefore seeks an optimum distance range for each direction, one which will allow the highest frequency of observations of high concentrations (or the highest exposure) to be monitored, subject to the constraints listed earlier. The principal ranking method used in MONITOR considers the frequency of meteorological events causing the plume to impact in the vicinity of the source. Subject to the constraints imposed by the concentration threshold, the resolution of peaks desired, and the specific weather classes included in the analysis, the ranking is based on the frequency with which an interval of downwind distance would receive the impact of the pollutant plume (see Figure 1). The exposure, which is simply the product of each peak concentration and its frequency, may be used as an alternate method, especially for siting particulate monitors. When considering cost efficiency, an inherent priority in network design, the minimum frequency of high concentration measurements needed to justify a monitor site must be identified. The statistical variation of rare meteorological events makes it inadvisable to locate a monitor in a location that is expected to experience the highest concentrations for fewer than 10 hr/year, unless that monitor also serves other purposes such as measuring background levels, distinguishing between two sources, or measuring high concentrations for another important averaging period as well. The MONITOR program itself is designed primarily for analysis of hourly concentrations. With appropriate power law averaging time factor adjustments to concentrations, shorter or longer periods up to 3 hr may be reasonably estimated. For 3, 8, and 24 hr period impact analysis, sequential modeling is replacing statistical methods as the technique favored by U.S. EPA. For that reason, the supplemental analysis of sequential model results is generally undertaken. For 3 hr periods, the statistical method, with adjustment for averaging time, may still be preferable because the use of sequential models to define locations of maximum concentrations is more costly. Sequential Modeling Method
In order to identify monitor sites appropriate for assessing compliance for longer than 1 hr averaging times, one of ERT's sequential models (i;e., MPSDM, PSDM5, or ERT-RAM) is often used in conjunction with the statistical processing subprograms (CUMFREQ and TOPTEN). CUMFREQ calculates cumulative frequency distributions at each receptor for any averaging period of interest. TOPTEN identifies the highest 10 concentrations at each receptor for any averaging period of interest. Another subroutine reports the meteorological conditions associated with these maxima, so that these cases can be checked for representativeness of observational data. CUMFREQ output can be analyzed to determine areas that have similar frequencies for selected concentration thresholds and specified averaging periods. CUMFREQ can also be used to evaluate the annual averages at each receptor to identify areas of highest annual impacts. The sequential method is more costly than MONITOR, even when limited to one or two years of data. The sequential method does offer the advantage that it allows use of exact meteorological observations instead of less precisely classified 712
As suggested above, the use of a geographical coverage model represents a generalized approach to the design of a network, and theoretically could make use of either of the methods described above. Hougland and Stephens,5 using a grid of receptor zones, gives for "monitor oriented" coverage factors Z; for site ;': i
k
and
ijk = FREQ(/e) • STR(i) • (1/1 + Dtj) where Zj A^k FREQ(&) STR(t') Dij
is monitor / coverage factor is coverage for each case condition is frequency of downwind direction k is source strength for source i is dilution factor for monitor j with respect to source i The MONITOR method also calculates, at user option, the Aijk values (and by tabular accumulation the Z, values) but MONITOR replaces STR(i). (1/1 + Du) with a gaussian diffusion model concentration for each contributing source. Miller and Noll4 have demonstrated in detail the utility of the coverage model concept, using a gaussian point source model as the basic method for evaluating concentration profiles. They demonstrate how, with other supplemental statistical subroutines, confidence levels as well as probability of capture can be estimated from the same basic analysis. Hougland and Stephens,5 in applying the coverage method to a complex urban area, found it advisable to consider a modification that prevented large sources from completely dominating the choice of monitor sites. Hougland and Stephens limited choices of monitors to those which had the largest geographical coverage factor for each source and wind direction. This maximized the overall geographical coverage of the network, allowing the effects of both the largest and smaller sources to be monitored with the final nine-station network. Number of Monitors
It must be apparent that the number of monitors required is also determined by a variety of site-specific factors, including: • remoteness of the proposed or existing source from other sources • regional background air quality status • number of major sources • complexity of meteorological flow in the area • number of pollutants that must be monitored • averaging times of standards appropriate for the most important pollutants • number of purposes to be served by each monitor The EPA guidelines for PSD monitoring suggest that the minimum number of baseline monitoring sites ranges from one to four, depending on the remoteness of the source and the relative impact of each source pollutant. For complex urban situations or in mountainous terrain, two to four times as many sensors may be needed to identify the contributions of the several major sources, or to cover a wide range of meteorological cases. Due to model limitations, the quantitative Journal of fhe Air Pollution Control Association
methods discussed above do not provide exact values for the frequencies of above-threshold concentrations detected, but they do allow the relative values for alternative network designs to be compared objectively. These quantitative methods also provide estimates of the total risk of capturing (or of missing) an important concentration level when the final selections of network sites have been made.
Downloaded by [190.42.216.132] at 11:37 30 January 2016
Practical Considerations
It should be noted that the MONITOR program optimizes its recommended sites strictly on the basis of the concentration and frequency ranking scheme selected. It is often necessary to exclude a number of the preferred sites for practical reasons such as (1) inaccessibility, (2) lack of power, (3) inadequate security, and (4) unavailable private property. Other qualitative factors that may favor a particular site that is ranked slightly lower by the MONITOR program are (1) protection of sensitive receptors, (2) control of maintenance costs, or (3) public relations aspects. Preparation of overlay maps of the recommended zones allows rapid consideration of these factors when these overlays are superimposed on an appropriately annotated land use map. Alternatively, the model used for calculations may be provided with an array of receptor-specific weighting factors that may be completely controlled by the user—another application of the coverage model concept.
Table I. Sample monitor results. (Threshold = 365 jig/m3; Resolution = 80%) Rank
Direction
Range (km)
% Frequency
%> Threshold
1 2 3 4 5
SSW SW S NVV SW
12.0-18.7 12.0-18.7 12.0-18.7 12.0-18.7 4.4- 5.0
2.59 2.56 2.00 1.70 1.64
3.96 4.74 3.62 3.28 4.74
Table II. Sample monitor results. (Threshold = 650 Mg/ Resolution = 80%) Rank 1 2 3 4 5
Direction SW WSW SSW ENE S
Range (km)
% Frequency
%> Threshold
4.3-5.0 4.3-5.0 4.3-5.0 7.3-7.4 4.3-5.0
1.60 1.00 0.96 0.86 0,81
1.95 1.30 1.11 1.12 1.08
Frequency > threshold for 2400 classes = 15.2
fraction of the concentrations above the threshold in a wind sector, which will be successfully detected by a monitor located within the specified distance range in that sector. The quotient of the % frequency divided by the % frequency exceeding the threshold is a measure of the monitor's efficiency in detecting peaks. Network Design
The culmination of any monitor siting analysis is the design of the final recommended network. This often requires com-, bining results of several MONITOR analyses and sequential modeling to derive a set of siting zones. Figure 2 shows a composite diagram of a monitoring network designed for an urban area containing three major SO2 sources. The final monitoring network design ranked most highly those zones that showed coincidence of areas predicted to have high annual averages or frequently high 3 hr concentrations with those locations recommended by MONITOR.
A SOURCE
A
c SOURCE
% Frequency > threshold for 2400 classes = 41.6
Sample of MONITOR Results
To illustrate the information available from the MONITOR program, two examples are given in Tables I and II. The major difference between the two sample analyses is the change in the specification of the threshold concentration. At the foot of each table, the total frequencies (hr/year) of cases exceeding threshold is given, as well as the number of weather classes considered. Both analyses used a resolution of 80% of maximum concentration. Selections were ranked by frequency with criteria for both tables. The first table resulted from a threshold of 365 ixg/m-\ equal to the 24 hour SO2 standard. The 650 ng/rri3 threshold is one-half the secondary 3 hour SO2 standard. In the tables, the first column on the left lists the rank, followed by the wind direction sector, and downwind ranges (xi and X2), which delimit the zone recommended for a monitor site that will meet the constraints of the particular analysis. Comparison of Tables I and II shows that the primary site recommendations move closer to the source as the threshold is raised. This is consistent with the fact that slightly unstable conditions with relatively high wind speeds yielded higher maximum concentrations than the higher wind speed neutral cases associated with the recommended distance range of 12 to 18.7 km. The next two columns can be used to identify the "coverage," as well as the measurement frequency associated with each rank. The sector-specific coverage is defined as the July 1979
Volume 29, No. 7
Area of Maximum Annual Average SO^ Area of 3 Hour Average SO2 in Excess of Threshold Optimum Sites Selected by Monitor Figure 2. Basis for recommended monitoring zones for three major sources in an urban area.
Figure 3 illustrates another situation in which predictions of the MONITOR program were compared to the zones recommended by the application of the EPA CRSTER program. The maximum and second highest individual concentrations predicted by CRSTER and the critical receptors identified by the EPA RAM model were also considered in the composite analysis shown. In each case it was relatively easy to identify single monitor sites that "covered" more than one critical zone in which a concentration near the relevant NAAQS level might occur frequently enough to be detected. Tables I and II have shown the decrease in coverage that occurs with establishment of a higher threshold. The cumulative coverage is analogous to an annual average concentration in that it has its frequency normalized to an annual basis. Thus, a cumulative value for coverage of 0.05 means that 5% of the year (438 hr) will be "covered" by the set of monitors in the specified locations. The cumulative value of the fre713
Downloaded by [190.42.216.132] at 11:37 30 January 2016
quency for all of the monitors chosen for a network should be at least 5 to 10% to ensure a useful quantity of plume impact data, and efficient use of the instrumentation. A network containing only sites selected on the basis of maximal thresholds will be very low in coverage, and therefore low in cost efficiency. This conflict may force a reevaluation of original strategy priorities. A compromise is placing the emphasis in design strategy on calibration of validated air quality models, instead of relying on direct measurement of the rare events to verify compliance (or prove noncompliance). The number of monitors required for actual validation of a model is quite large, requiring many
\\.*-J*
climatological frequency of occurrence of the concentrations expected in the vicinity of a set of sources. The siting method requires that a set of monitoring priorities be identified. Several possible strategies for accomplishing this have been suggested. The analysis methodology for a climatological statistical model, MONITOR, has, been described. The value of the frequency provided has been shown with a sample analysis. The opportunities for the analyst to specify qualitative weighting factors, or to use additional supplemental models for assessing impacts for averaging times longer than an hour have also been illustrated. The goal is to combine all of these factors into a single network design. Even if PSD requirements motivate the installation of a network, the other potential criteria for siting monitors should be considered. The resulting networks will then provide an efficient and effective information system that meets the requirements defined in the monitoring strategy. The potential conflict between network cost-efficiency and a strategy that requires direct measurements of compliance with NAAQS has led to the following conclusion. Unless a large number of sites can be instrumented, a monitoring network should be designed to measure concentrations significantly higher than background, but with a sufficient measurement frequency to provide a good basis for comparison with air quality model predictions. This approach will be more successful than one which attempts to provide direct measurements of the absolute maxima. Acknowledgments
The authors have been assisted by the early development work of Mr. J. P. Gaertner on the MONITOR program. Dr. J. Nuber's implementation of several program modifications is gratefully acknowledged. Mr. R. V. Bibbo is also thanked for his suggestions and recommendations for supplemental application of sequential modeling methods. References
KEY •
Emission Source
A
Existing State Monitor
— — — CRSTER Zones —
MONITOR Zones
*
•
RAM Critical Receptors
•X"
CRSTER-Estimated Highest and HighestSecond Highest 24-Hour Concentrations
Figure 3.
Multiple criteria for monitoring site selection for two rural
1. "Ambient Monitoring Guidelines for Prevention of Significant Deterioration," U.S. EPA, EPA-450/2-78-019 (OAQPS No. 1.2096), 1978. 2. Federal Register, 43:118(1978). 3. D. G. Smith, "A Quantitative Method for Selecting Effective Air Quality Monitoring Sites," presented at the 71st Annual Meeting of the Air Pollution Control Association, Houston, TX, June 25-29 1978. 4. T. L. Miller and K. E. Noll, "Design of Air Monitoring Surveys near Large Power Plants," in Power Generation: Air Pollution Monitoring and Control, Ann Arbor Science Publishers, Inc., Ann Arbor, MI, 1976. 5. E. S. Hougland and N. T. Stephens, "Air pollutant monitor siting by analytical techniques," J. Air Poll. Control Assoc, 26: 51 (1976). 6. K. E. Noll, T. L. Miller, J. E. Norco, and R. K. Raufer, "An objective air monitoring site selection methodology for large point sources," Atmos. Environ., 11:1051 (1977).
monitors aligned on acres about the source, as can be seen by considering the refined statistical analysis model developed by Noll, et al.6 Once a model is validated, however, its calibration can be refined by statistical analysis of easily measured concentration levels which occur with a high frequency. In this way, confidence in the precision of maximum values predicted by mathematical models can be gradually improved. This confidence is most important when the extremely low concentrations associated with PSD increments cannot be measured. Summary and Conclusions
This paper has discussed quantitative methods for objectively ranking potential monitoring sites on the basis of the 714
Mr. Smith and Mr. Egan are with the Air Quality Center, Environmental Research & Technology, Inc., 3 Militia Drive, Lexington, MA 02173.
Journal of the Air Pollution Control Association