Dispersion Modeling Guidance for Airports Addressing Local Air Quality Health Concerns (2017)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

20 The general objective of the intercomparison was to use equivalent input data in all four models, keeping the inputs consistent so that any differences in output would be the result of dif- ferences in the dispersion models. Of the four models, only AERMOD could directly support the nearly 5,000 area sources to represent aircraft activity; the other models needed to be custom- ized for this application. For example, CALPUFF had to be modified to increase the number of area-source puffs from 200 to 6,000, and subsequently the total number of puffs from 100,000 to 500,000. Similarly, SCICHEM hit CPU-memory limits with the number of puffs that were gen- erated with approximately 5,000 area sources. To compensate for this, the aircraft sources were split into 4 subsets, and thus required four separate executions each time. Although AERMOD and CALPUFF continue to keep the same dimension for the area source during their modeling of aircraft sources, SCICHEM splits this source into 10 smaller Gaussian points. By doing this for the LAX study, the researchers found that the total number of sources increases by about a factor of 10. This increase is the key reason behind the excessive run times associated with SCICHEM modeling of aircraft emissions as area sources. There were 4,179 aircraft sources whose aspect ratio was 10.0; hence, SCICHEM split each of the 4,179 sources into 10 Gaussian point sources, resulting in a total of 41,790 sources. Thus, what started as 6,170 sources in EDMS/AERMOD were modeled as 52,035 sources in SCICHEM. To support the model evaluation and intercomparison, the researchers configured all the models to predict concentrations at the LAWA AQSAS measurement locations. The research team also set the AERMOD and CALPUFF models to predict pollutant concentrations for a uni- form Cartesian grid centered on the airport for a 5 × 5 km region (with receptors every 500 m), a Polar grid centered on the airport for a 50 × 50 km region (with receptors every 5 km), and flag- pole receptors aloft to capture vertical gradients. Based upon computational demands of some models, however, this expanded set of receptors was used only with AERMOD and CALPUFF, and not with the other two models. All models were configured to predict seven pollutants: CO, NOx, SO2, VOCs, TOG, PM2.5 and PM10 on an hourly basis for each of the two 6-week peri- ods. AERMOD, SCICHEM, and CALPUFF were configured with most of their default options; ADMS-Airport was used with the Air File (ADMS-AIR) input option, and all results have been presented for this combination of four models, unless otherwise stated. In the cases of CALPUFF, SCICHEM, and ADMS-Airport, the researchers performed addi- tional sensitivity simulations to take advantage of specific enhancements that each model offered to provide improved characterization of local-scale air quality at the airport. The simulation scenarios were: 1. CALPUFF Sensitivity #1 with the Slug option for aircraft sources, 2. CALPUFF Sensitivity #2 with CALMET-based meteorological inputs, 3. SCICHEM Sensitivity #1 with reduced number of sources, C h a p t e r 5 Dispersion Model Intercomparison

Dispersion Model Intercomparison 21 4. SCICHEM Sensitivity #2 with full chemistry, and 5. ADMS-Airport Sensitivity #1 with Volume source option for aircraft sources (ADMS-VOL). The model outputs were compared against observations at the 17 LAWA sites, as well as against each other, using an extensive set of graphical and statistical measures of model perfor- mance (including mean bias, mean error, fractional bias, factor of 2, normalized mean square error [NMSE] and geometric mean bias). The next sections in this chapter provide a brief summary of the main results, followed by a discussion of their meaning with respect to problems with the dispersion models. Figure 3 presents a summary of model performance for NOx at the four LAWA core sites compared to observations, as a function of wind speed and wind direction. All models overpredict observed concentrations at CE, the site located east of the South runway and the site with the highest observed concentrations of NOx. The overprediction in CALPUFF was the highest, with AERMOD closest to the observed concentrations. Similar comparisons are presented for PM2.5 in Figure 4. Figure 3. Comparisons among models (model-to-model) and with observed NOX concentrations for summer 2012.

22 Dispersion Modeling Guidance for airports addressing Local air Quality health Concerns All models underpredict observed PM2.5 at all four sites, which highlights the lack of background concentrations in the local-scale modeling, thus pointing to the need to incorporate background concentrations using another regional-scale model like the CMAQ model, or to use geostatisti- cal techniques such as Space-Time Ordinary Kriging (STOK) of observed concentrations from remote background locations. The performance results show that the highest observed and predicted concentrations of NOx usually occur at night, typically after 8:00 p.m. (stable conditions) with the minimum during daytime, from 10:00 a.m. to 6:00 p.m. Additional maxima cluster around night-to-day and day-to-night transition periods (i.e., from 6:00–8:00 a.m. and from 7:00–8:00 p.m.). With respect to atmospheric stability, as defined by the inverse Monin-Obukhov length (1/L), the highest concentrations usually occur under near-neutral conditions (1/L ~ 0), which happen at or near the transition periods, an observation that is consistent with the time-of-day results. High concentrations occur for both unstable conditions (daytime, 1/L < 0) and stable condi- tions (1/L > 0); that is, they exist on both sides of the neutral stability limit. Furthermore, the Figure 4. Comparisons among models and with observed PM2.5 concentrations for summer 2012.

Dispersion Model Intercomparison 23 highest concentrations occur under light winds (1–3 m/s typically) for both stable and unstable conditions. Based upon the specific pollutant that is being considered, and how it is evaluated according to the NAAQS (i.e., whether the focus is on a short-term maximum concentration or on a longer-term average concentration), these patterns with model performance have implica- tions for airport-related local air quality health impacts. The models generally overestimate the highest concentrations, as shown by quantile-quantile (Q-Q) and box-and-whisker plots (shown in the contractor’s final report, available online), and the largest statistical variability or highest concentration fluctuations are associated with the maximum concentrations. This pattern points to problems of plume transport and disper- sion under light wind and stable atmospheric conditions, which have been found previously for AERMOD and other models (Cirillo and Poli 1992; Sharan and Yadav 1998). Similar light wind problems occur under convective conditions with these models (Weil, Corio, and Brower 1997). Given the number of pollutants that were modeled at different sites, during two different seasons, the researchers developed a simple objective scoring scheme to group all model results into one of three bins: “Good” (modeled mean between the 25th and 75th percentile of observed means), “High” (modeled mean above the 75th percentile of observed means) and “Low” (modeled mean less than the 25th percentile of observed means). The resulting predictions for NOx, CO and SOX were combined. One point was awarded for each combination of model/site/pollutant if the result fell in the appropriate category of Good, High, or Low. Half a point was awarded if the model result fell in the borderline region between any two categories. The researchers excluded PM2.5 from this analysis because all models underestimated PM2.5 levels. The resulting scores appear in Table 4 and Table 5 in relation to two specific metrics of model performance. To compute the scores in Table 4, the researchers looked at the ranges of hourly concentrations predicted by each model and compared them to the observed ranges. Each model was eligible to score a maximum of 24 points (6 pollutants at the 4 LAWA core sites). Ideally, the model providing results that most closely matched observed conditions would have 24 points under the Good category. High Low Good AERMOD 1 5 CALPUFF 3 3 SCICHEM 2 4 ADMS-Airport 1 5 Table 5. Model performance objective scores, based upon NMSE vs. FB for NOX, CO and SOX. High Low Good AERMOD 3 6 15 CALPUFF 12 3 9 SCICHEM 3.5 3 17.5 ADMS-Airport 7 4 13 Table 4. Model performance objective scores, based upon observed and modeled distributions of NOX, CO, and SOX.

24 Dispersion Modeling Guidance for airports addressing Local air Quality health Concerns Based upon this scoring scheme, in the Good category SCICHEM—with a score of 17.5— seems to slightly outperform AERMOD with a score of 15 and ADMS-Airport with a score of 13. CALPUFF has 12 points under the High category, with only 9 under the Good category. Table 5 shows a scatter of normalized mean square error (NMSE) versus fractional bias (FB) in making the same determination. Each model was eligible to score a total of 6 points. Based upon this metric, both AERMOD and ADMS-Airport—each with a score of 5 in the Good category— seem to slightly outperform the other two models. Using this methodology, CALPUFF scored 3 in both the High and Good categories. The following conclusions were drawn from the model intercomparisons for the LAX AQSAS: • Modeling systems and input datasets – EDMS-based emissions inventories are not directly usable by dispersion models other than AERMOD, and significant efforts need to be devoted to this task. – AERMET-based meteorology for non-steady-state models such as CALPUFF and SCIPUFF needs to be reviewed carefully to ensure continuous hours with valid meteorological data. – The EDMS approach to modeling area sources does not translate well to SCICHEM, caus- ing significantly longer runtimes, for approximately 5,000 sources. Additional work is needed to aggregate sources. – Key air pollutants of interest from a health risk point of view are fine particulate matter (i.e., PM2.5), followed by O3, and then air toxics to a relatively smaller degree. – The researchers focused the model intercomparison on NOx, PM2.5, CO, and SO2 as key pol- lutants of interest. Given that O3 is a secondary pollutant, not all of the four chosen models could predict it. Nevertheless, O3 is not formed appreciably in the immediate vicinity of the airport, due to rapid titration by high levels of NOx by aircraft. • Model predictions – Models tend to overpredict summer NOx but underpredict winter NOx levels. NOx measured during winter (55 - 100 µg/m3) was higher than NOx measured during summer (7.5 - 35 µg/m3). – The AERMOD- and SCICHEM-predicted means are closer to observations, whereas ADMS-Airport and CALPUFF tend to overpredict. – Compared to the other models, CALPUFF predicts the highest contributions from aircraft sources. – AERMOD- and SCICHEM-predicted distributions are closer to observed than CALPUFF and ADMS-Airport. – PM2.5 is a criteria pollutant of special interest in relation to airports’ local air quality and health impacts. Predictions of PM2.5 are poor across-the-board, pointing to lack of back- ground concentrations, and hence secondary components of PM2.5. ▪ A need exists to incorporate regional background concentrations using hybrid techniques. ▪ Future research could focus on local-scale models that can incorporate this process with- out substantially affecting model runtimes. • Maximum concentrations are overpredicted by AERMOD, possibly because of missing plume rise, but means are reasonably predicted. This result shows the potential for AERMOD and other models to be conservative in application to short-term maximum concentrations (such as the 1-hour form of the NO2 NAAQS, which requires computation of the 98th percentile of the 1-hour daily maximum averaged over 3 years), but reasonable for predicting long-term concentrations (such as the annual average form of the NO2 NAAQS). • Both CALPUFF and ADMS-Airport show much larger contributions from non-aircraft sources, highlighting potential differences in treatment of aircraft sources. • UFP was not modeled, because the EDMS inventories did not support them. Given evolving literature on airport-based UFP studies, EDMS or AEDT could be enhanced to generate emis- sions of UFP for activities at airports.