Grantee Research Project Results
Final Report: Air Pollutant Concentrations in the Vicinity of Buildings: Model Development and Evaluation
EPA Grant Number: R826156Title: Air Pollutant Concentrations in the Vicinity of Buildings: Model Development and Evaluation
Investigators: Peterson, Holly G. , Lamb, Brian , Stock, David , Schulman, Lloyd
Institution: Montana Tech of the University of Montana , Washington State University
EPA Project Officer: Hahn, Intaek
Project Period: October 1, 1997 through September 30, 2000
Project Amount: $360,011
RFA: Exploratory Research - Physics (1997) RFA Text | Recipients Lists
Research Category: Water , Land and Waste Management , Air , Safer Chemicals
Objective:
The current version of the Industrial Source Complex (ISC) model contains empirical algorithms for plume downwash, but a (well-mixed) cavity concentration is predicted only via simple assumptions in the screening model, SCREEN. To improve our ability to predict concentrations in the vicinity of buildings, the Electric Power Research Institute (EPRI), in cooperation with the Environmental Protection Agency (EPA), recently sponsored a project to develop the Plume Rise Model Enhancement (PRIME). The new ISC-PRIME codes explicitly predict the trajectory of a plume, and they include refined downwash algorithms in addition to realistic calculations for cavity concentrations. Output of ISC-PRIME consists of concentrations downwind of pollutant sources with averaging times between 1 hour and 1 year, including concentrations near buildings in recirculation cavities and downwash regions. Limited field measurements, however, have been available to test performance of ISC-PRIME. Furthermore, very few data exist to examine concentration fluctuations that occur on time scales much shorter than 1 hour from unsteady winds and from intermittent cavity development.
Overall goals of this research project were to evaluate ISC-PRIME with field data and to improve our fundamental understanding of plume dispersion near buildings. Specific objectives were to: (1) conduct tracer experiments in the vicinity of isolated buildings under a range of meteorological conditions; (2) compare field results to model predictions from ISC-PRIME; (3) investigate concentration fluctuations in and near cavity regions; and (4) employ a numerical modeling approach to aid interpretation of the results.
Experimental Approach. Plume behavior was investigated for this project in the vicinity of isolated buildings during three field campaigns: (1) the Fairgrounds Study, (2) the Icefield Study, and (3) the Skating Center Study. Terrain included open fields and a smooth ice surface, while meteorological conditions encompassed stable through unstable categories. Building types ranged from a small office trailer to an industrial-sized structure, and winds were monitored onsite. In all experiments, sulfur hexafluoride (SF6) gas was released at a constant rate from locations within the recirculation cavity, at rooftop, or above the building. Tracer concentrations were measured near ground level with arrays of time-averaged samplers and with fast-response analyzers.
Summary/Accomplishments (Outputs/Outcomes):
Model Testing. Current versions of ISCST3 (version 99155), SCREEN3 (version 96043), and ISC3-PRIME (version 99020) were used for model testing with the Fairgrounds data. Tables 1 and 2 summarize model performance statistics for the Fairgrounds data in terms of predicted-to-observed ratios, factor of 2 and factor of 3 analyses, and an overall skill measure in the form of an absolute value fractional bias ratio (AFBR). The AFBR is equal to ΣAFB/ΣAFBmax, or the ratio of the summation of absolute value fractional biases to the summation of absolute value fractional biases for a model with no skill. The ΣAFBmax term is equivalent to 2 times the number of comparison points and is the outcome that would result for a model with no predictive skill. By summing the AFBs for all sample points in a group and then dividing by 2 times the number of comparison points, the predictive capability of a model is scaled between 0 and 1.0. Thus, for a model with perfect predictive skill, the AFBR (expressed as a percent) would be 0 percent, and for a model with no skill the AFBR would be 100 percent.
Table 1. Fairgrounds Study—ISC3-SCREEN3 Statistics | ||||||
CATEGORY | # Points | P/O RANGE | P/O < 1 (%) | FAC2 (%) | FAC3 (%) | AFBR (%) |
All points
| 135 | 0-55.1 | 57.0 | 30.4 | 46.7 | 50.9 |
SCREEN3 | 78 | 0.43-55.1 | 25.6 | 50.0 | 65.3 | 38.1 |
ISC3 | 57 | 0-0.55 | 100 | 3.5 | 21.1 | 68.4 |
Cavity
| 114 | 0-55.1 | 49.1 | 34.2 | 46.5 | 50.3 |
Far wake | 21 | 0.02-0.67 | 100 | 9.5 | 47.6 | 54.2 |
D stability | 52
| 0-24.4 | 30.8 | 44.2 | 57.7 | 43.1 |
E stability | 65 | 0-55.1 | 70.8 | 21.5 | 40.0 | 56.9 |
F stability | 18 | 0.15-4.72 | 83.3 | 22.2 | 38.9 | 51.9 |
Row maxima | 43 | 0.09-7.65 | 69.8 | 25.6 | 37.2 | 41.0 |
Test maxima | 10 | 0.09-3.51 | 70.0 | 40.0 | 70.0 | 33.8 |
Windward release | 109 | 0-55.1 | 50.5 | 28.4 | 43.1 | 52.7 |
Leeward release | 26 | 0.06-9.19 | 91.7 | 38.5 | 61.5 | 43.4 |
1.0 Hb release | 99 | 0.06-24.4 | 60.6 | 41.4 | 61.6 | 40.6 |
1.5 Hb release | 36 | 0-55.1
| 47.2
| 0
| 5.6 | 79.1
|
# Points = Number of nonfringe points; P/O Range = Range of predicted-to-observed ratios | ||||||
P/O < 1 = Percentage of points underpredicted by model | ||||||
FAC2 = Percentage with a P/O ratio within a factor of 2; FAC3 = within a factor of 3 | ||||||
AFBR = Absolute Fractional Bias Ratio (AFBR = 0% for a model with perfect performance, and AFBR = 100% for a model with no skill) |
Based on Tables 1 and 2 for the Fairgrounds study, Dunleavy (2000) summarized the following results concerning relative performance of the two models:
• Performance of ISC3-PRIME was superior to that of ISC3-SCREEN3 for all categories and all statistics except for the percentage of points underpredicted.
• The ISC3-PRIME model was less conservative than SCREEN3 in its prediction of cavity concentrations for most tests. In particular, ISC3-PRIME underpredicted 73.1 percent of its points, while SCREEN3 underpredicted 25.6 percent of the same points. This resulted in less underprediction by the ISC3-SCREEN3 model for most statistical analysis categories. Cases where SCREEN3 underpredicted concentrations were associated with rooftop releases with wind speeds less than 2.5 ms-1.
• Regarding row and test maximum concentrations, ISC3-PRIME underpredicted 93 percent and 100 percent of the time, respectively, while ISC3-SCREEN3 underpredicted approximately 70 percent of the time. However, ISC3-PRIME prediction was within a factor of 2 approximately 70 percent of the time for these categories, and ISC3-SCREEN3 was only within a factor of 2 for 25 to 40 percent of the time.
• The ISC3-PRIME model predicted 1.6 to 16.5 times more within a factor of 2, and 1.4 to 4.2 times more points within a factor of 3, than did the ISC3-SCREEN3 combination.
• The AFBRs for the ISC3-PRIME model ranged from 21.6 to 35.9 percent, while AFBRs for ISC3-SCREEN3 ranged from 33.8 to 79.1 percent. Again, the smaller the AFBR, the better the model performance.
Table 2. Fairgrounds Study—ISC3-PRIME Statistics. | ||||||
CATEGORY | # Points | P/O RANGE | P/O < 1 (%) | FAC2 (%) | FAC3 (%) | AFBR (%) |
All points
| 135 | 0.08-15.8 | 80.7 | 71.9 | 89.6 | 26.1 |
SCREEN3 | 78 | 0.21-6.84 | 73.1 | 82.1 | 89.7 | 21.6 |
ISC3 | 57 | 0.08-15.8 | 91.2 | 57.9 | 89.5 | 32.3 |
Cavity
| 114 | 0.21-15.8
| 77.2 | 72.8 | 91.2 | 24.3 |
Far wake | 21 | 0.08-0.82 | 100 | 66.7 | 90.0 | 35.9
|
D stability | 52
| 0.27-15.8 | 69.2 | 75.0 | 90.4 | 24.7 |
E stability | 65 | 0.08-3.7 | 87.7 | 69.2 | 89.2 | 26.1 |
F stability | 18 | 0.12-1.7 | 88.9 | 72.2 | 88.9 | 30.2 |
Row maxima | 43 | 0.21-1.3 | 93.0 | 76.7 | 95.3 | 23.7 |
Test maxima | 10 | 0.21-0.71 | 100 | 70.0 | 90.0 | 29.8 |
Windward release | 109 | 0.08-15.8 | 78.9 | 73.4 | 89.0 | 26.0 |
Leeward release | 26 | 0.32-3.41 | 88.5 | 65.4 | 92.3 | 26.0 |
1.0 Hb release | 99 | 0.08-6.8 | 83.8 | 69.7 | 89.9 | 27.3 |
1.5 Hb release | 36 | 0.21-15.8 | 72.2 | 77.8 | 88.9 | 22.8 |
Lacourciere (2001) tested ISC3-PRIME performance using the Fairgrounds data in 30-minute time blocks instead of 1-hour periods, and found similar results as Dunleavy in terms of a tendency to underpredict concentrations. Lacourciere's statistical analysis indicated an average predicted-to-observed (P/O) ratio of approximately 0.92, with about 71 percent of the ratios less than 1.0 and about 29 percent greater than 1.0. Again, PRIME appeared to perform equally well in stable and neutral conditions.
Lacourciere, however, found very different results testing ISC-PRIME with the Icefield dataset. In this case, the average P/O ratio was 3.42, with only 10 percent of the ratios less than 1.0 and 90 percent greater than 1.0.
These findings demonstrate the need to "use caution" when applying data from a single field site to test a diffusion model, even if a range of stability conditions are represented. Both Dunleavy and Lacourciere completed sensitivity analyses, and found ISC3-PRIME to be very insensitive to stability category regarding concentrations in the recirculation cavity. The model, however, tended to underpredict during stable conditions at the fairgrounds and overpredict during stable conditions on the icefield. While building size was similar in both studies, wind speeds were higher and vertical temperature gradients were stronger on the icefield.
Summary for Concentration Fluctuations. While time-averaged concentrations were addressed in the model-testing phase of this project, plume behavior also was analyzed in terms of time-scales reflecting human breathing rates. All pollutants do not cause adverse health effects on short time scales, but it is important to realize that "time-averaged" exposure (from field measurements or from diffusion models) may underestimate the risks for some contaminants that can cause acute respiratory responses within seconds. Neither air quality standards nor regulatory models have adequately addressed near-instantaneous plume behavior downwind of industrial sources, and this topic may be particularly consequential in the vicinity of buildings where pollutants accumulate in the cavity zones.
Our field data illustrated dramatic variability of real-time concentrations within the recirculation cavity. Figure 1 contains an example time series of 1 Hz data collected within the near wake during the Fairgrounds Study. For these data, the short-term peak concentration is approximately 6 times higher than the time-averaged concentration. Crosswind profiles of peak-to-mean (P/M) ratios showed P/M ratios near 2-4 at the center of the cavity, and up to about 12 near the fringes. Thus, even a "perfect" time-averaged model will underestimate short-term peaks in the cavity by, at minimum, a factor of 2, and possibly by a factor of 10 or more.
Figure 1. Time series of concentration fluctuations during one experiment of the Fairgrounds Study. Statistics for this time series include a mean of 370 ppt, a peak of 2218 ppt, an intensity of 0.76, an intermittency factor of 1, and a P/M ratio of 6.
Summary for Numerical Analyses. Finally, regarding the numerical objective of this project, few studies in the literature illustrate k-ε modeling techniques to predict flow about three-dimensional buildings in stable atmospheres. Therefore, for this project, flow was modeled using a fourth-order accurate finite element scheme. The time-averaged Navier-Stokes equations were closed with the standard k-ε turbulence model as well as with a k-ε- turbulence model, and dispersion results were compared with field measurements from the Icefield Study.
As presented by Brzoska, et al. (2000), ground-level tracer measurements for one of the Icefield experiments are shown in Figure 2. Corresponding concentrations predicted by the k-ε-model are illustrated in Figure 3, and by the k-ε model in Figures 4 and 5.
Highest concentrations in the k-ε-model occur at approximately one-quarter building height in the lee of the building with some transport along the side of the building. Very little impact is seen just behind the building, but a fan of plume material trails out behind the building. Experimental results indicate lower concentrations in the far wake region, but the sample array was too limited to make any quantitative comparison. The k-ε approach predicted higher concentrations in the wake of the building and along the outside of the building than did the k-ε- model. Figure 4 shows a vertical concentration profile from the k-ε model, and it is evident that ground-level values are a small fraction of concentrations nearer to the release. Overall, these results illustrate the strength of combining field measurements, which often cover a limited spatial array, with numerical methods to improve our understanding of plume behavior near an obstacle such as a building.
Figure 2. A top view of ground-level tracer concentrations. Concentration units are parts-per-billion (ppb).
Figure 3. A top view of the ground-level concentration profile from the k-ε-model. Again, concentration units are parts-per-billion (ppb).
Figure 4. A top view of the ground-level concentration profile from the k-ε model.
Figure 5. A side view of the concentration profile from the k-ε model.
Journal Articles:
No journal articles submitted with this report: View all 9 publications for this projectSupplemental Keywords:
air, ambient air, modeling, tracer experiments, building effects., Scientific Discipline, Air, Physics, Environmental Chemistry, Atmospheric Sciences, Engineering, Chemistry, & Physics, building vicinity, aerosol particles, air pollution concentrations, building plume downwash models, database development, air pollution, atmospheric stability, plume dispersion model, tracer experimentProgress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.