Grantee Research Project Results
Final Report: Quantification of Uncertainty in Air Quality Models Used for Analysis of Ozone Control Strategies
EPA Grant Number: R824792Title: Quantification of Uncertainty in Air Quality Models Used for Analysis of Ozone Control Strategies
Investigators:
Institution: University of Colorado at Boulder , University of California - Berkeley
EPA Project Officer: Chung, Serena
Project Period: October 1, 1995 through September 30, 1998 (Extended to September 30, 1999)
Project Amount: $426,233
RFA: Air Pollutants (1995) RFA Text | Recipients Lists
Research Category: Air Quality and Air Toxics , Air
Objective:
Uncertainties in models used to predict the air quality impacts of prospective control measures have been cited as one reason why many urban areas have not been able to attain the National Ambient Air Quality Standard for ozone. Because air quality models play a central role in the design of strategies for reducing urban ozone, it is critical that comprehensive, systematic uncertainty analysis of these models be undertaken. The objectives of this project were to (1) demonstrate the use of systematic methods for uncertainty quantification and indentification of influential sources of uncertainty; (2) develop new approaches for estimating uncertainties in model parameters and inputs; and (3) explore how modeling uncertainties affect the assessment of air quality impacts of control measures. To meet these objectives, the study utilized the California/Carnegie Institute of Technology (CIT) airshed and trajectory models, applied for simulation conditions in the Los Angeles area on August 27-28, 1987. A quantitative uncertainty analysis has been completed for the CIT trajectory model (Bergin et al., 1999) accounting for uncertainties in more than 50 model inputs or parameters including emissions, wind fields and resulting trajectory paths, chemical parameters, deposition affinities and mixing depths. In addition, the project also investigated a new uncertainty analysis technique, Bayesian Monte Carlo analysis, which combines subjective "prior" uncertainty estimates developed by standard Monte Carlo techniques with information about the agreement between model outputs and observations (Bergin and Milford, 1999).A key component of the project has been the development of new approaches for estimating uncertainties in critical model parameters and inputs. This aspect of the study included a detailed exploration of uncertainties in windfields used as input to the trajectory model (Noblet et al., 1998) and the use of an alternative motor vehicle emissions inventory based on gasoline sales data and infrared remote sensing measurements (Harley et al., 1997). The assessment of emission inventory uncertainties has recently been extended to consider diesel-powered off-road mobile sources in the construction, agriculture, mining, and other sectors. The exploration of wind field uncertainties is being extended to the three-dimensional case in ongoing research that was started with support from this grant.
Summary/Accomplishments (Outputs/Outcomes):
Harley et al. (1997) investigated the effect of revisions to the motor vehicle emissions inventory and chemical mechanism on the performance of the CIT airshed model for the August 27-28, 1987 ozone episode in southern California. The SAPRC93 chemical mechanism used in the model was found to predict peak ozone concentrations that averaged 22% higher than the LCC mechanism used previously. A revised motor vehicle emission inventory was developed using gasoline sales and infrared remote sensing data for CO and measured ambient ratios of non-methane organic compounds (NMOC) to CO and nitrogen oxides (NOx) to CO. Using this approach, on-road vehicle NMOC and NOx emissions for the South Coast Air Basin in summer 1987 were estimated to be 2.4 and 1.0 times, respectively, the corresponding official inventory estimates from the MVEI 7G model. Ozone concentrations predicted using the CIT airshed model matched observations more closely when the revised inventory was used in place of the official estimates. The study also explored the hypothetical case of combining 1987 traffic volumes with precontrol vehicle emissions factors. If the vehicle fleet in 1987 were operating with no emission controls, the study found that NMOC and NOx emissions would have been 3.3 and 1.7 times higher, respectively, than the actual gasoline sales-based estimates for 1987. On average, peak predicted ozone concentrations for the controlled vehicle fleet were 43% lower than values predicted for the hypothetical uncontrolled fleet.In ongoing research, the assessment of emission inventory uncertainties described by Harley et al. (1997) has been extended to consider diesel-powered off-road mobile sources. The method used by EPA's Office of Mobile Sources, and coded in the new EPA NONROAD emissions model, estimates off-road engine activity based on the population (number) of engines in use, the estimated hours of use per engine per year, and the load factor (average load as a percent of maximum rated horsepower when running). This product gives total horsepower-hours of engine operation, and is multiplied by emission factors (grams emitted per hp-hr of engine activity) to calculate emissions. An alternate method for estimating emissions from these sources was developed based on fuel sales surveys conducted by the Energy Information Administration of the U.S. Department of Energy, using fuel sales rather than hours of engine operation as a measure of source activity. This avoids the uncertainties associated with estimating engine populations, hours of use, and load factors. We assessed emissions from off-road mobile sources, and found that total NOx emissions from off-road diesel engines nationwide (excluding railway locomotives and ships) were overstated in EPA's 1996 Emission Trends Report by a factor of about two.
Noblet et al. (1998) examined the effect of wind field uncertainty on the predictions of the trajectory version of the CIT model. Surface wind field uncertainties were quantified using data from the August 27-28, 1987 episode. A hybrid objective/diagnostic wind model was used to interpolate hourly wind speed and direction measurements at 56 monitoring sites to a regular two-dimensional grid system. Wind field uncertainty was quantified using a data withholding procedure. During daytime hours, the interpolation residuals were normally distributed, with mean values of ~0 and standard deviations of 0.83 and 0.64 m s-1 for the u and v wind components, respectively. During nighttime hours, residuals were normally distributed with mean values of ~0 and standard deviations of 0.43 and 0.35 m s-1 for u and v, respectively. The residuals were not spatially correlated throughout the air basin, but were found to be serially correlated. The daytime u and v serial autocorrelation coefficients were +0.50 and +0.49, respectively, while nighttime coefficients were +0.33 and +0.21.
Utilizing the wind field uncertainty estimates, Noblet et al. (1998) generated ensembles of 100 air parcel trajectories for air quality monitoring sites throughout the Los Angeles area. A stochastic component representing the uncertainty in the interpolation procedure was incorporated into the surface wind field at each hour using Latin hypercube sampling. Emissions data were accumulated for each air parcel trajectory for input to the CIT trajectory model. The model was then run to produce distributions of predicted pollutant concentrations over the ensemble of trajectories. Results for air parcels arriving at eight receptor sites on the afternoon of August 28, 1987 indicated an uncertainty (? one standard deviation relative to the mean) of 12 to 68% in predicted ozone concentrations due to wind field uncertainties. As mentioned above, the analysis of wind field uncertainty contributions is being extended to 3-D models in ongoing research.
The wind field uncertainty estimates developed by Noblet et al. (1998) and on-road motor vehicle emissions estimates of Harley et al. (1997) were included in a comprehensive uncertainty analysis of the CIT trajectory model that has been described by Bergin et al. (1999). This part of the study applied Monte Carlo analysis with Latin hypercube sampling to evaluate the effects of uncertainty in air parcel trajectory paths, emissions, rate constants, deposition affinities, mixing heights and atmospheric stability on predicted concentrations of ozone, formaldehyde, nitric acid, peroxy acetyl nitrate and total oxides of nitrogen (NOy). Uncertainties in predicted changes in ozone due to 25% reductions in motor vehicle NMOC and NOx emissions were also estimated.
Estimated uncertainties in ozone ranged from 24 to 57% (1s relative to the mean) across four Los Angeles area receptor sites. Seven variables contributed almost 80% of this uncertainty. These variables include motor vehicle emissions (uncertainty in both NMOC and NOx emissions due to the infrared remote sensing estimation procedure), the trajectory path (uncertainty due to wind field interpolation), and rate constants for HNO3 formation, NO2 photolysis, the acetyl peroxy radical (RCO3) + NO reaction and PAN decomposition. Reductions in motor vehicle NMOC were found to reduce ozone by 10 ? 10% to 28 ? 10%. With reductions in motor vehicle NOx emissions, the change in ozone ranged from an increase of 14 ? 14% to a decrease of 6.6 ? 6.2%. The major sources of uncertainty in the predicted responses to motor vehicle emissions reductions include the rate constants for HNO3 formation and the RCO3 + NO reaction, the trajectory paths and the NO2 deposition affinity.
A new uncertainty analysis technique, Bayesian Monte Carlo (BMC) analysis, was investigated by Bergin and Milford (1999). Bayesian Monte Carlo analysis provides a means of combining subjective "prior" uncertainty estimates developed by standard Monte Carlo techniques with information about the agreement between model outputs and observations. The resulting "posterior" uncertainty estimates reflect both the model's performance and subjective judgments about uncertainties in model parameters and inputs. To demonstrate the approach, BMC analysis was applied to the CIT trajectory model for two-day trajectories ending on August 28, 1987 at Azusa and Riverside, CA. Refined estimates of uncertainties in base case O3 concentrations were calculated, along with estimates of uncertainties in the response to 25% reductions in motor vehicle emissions of NOx and NMOC. For the cases studied, the model results were in reasonable agreement with spatially interpolated observations (Figure 1). Bayesian updating reduced the estimated uncertainty in predicted peak O3 concentrations from 35 to 20% at Azusa and from 24 to 18% at Riverside. At Azusa, the prior and posterior model results are consistent in indicating that reductions in NMOC emissions would be more effective than reductions in NOx emissions, which have a substantial probability of increasing final ozone concentrations. At Riverside, the mean of the prior distribution suggests that reducing NMOC emissions would be more effective than reducing NOx emissions, but the median (50th percentile) result suggests the opposite. Both the means and the median estimates from the posterior distributions show NMOC reductions as being more effective than NOx reductions.
Figure 1. Comparison of ozone concentrations from prior Monte Carlo modeling (Bergin et al., 1999) versus interpolated observations along the trajectory ending at Azusa, CA at 1:30 p.m. on August 28, 1987. The mean value and mean numbers are shown for each case. (From Bergin and Milford, 1999.)
Conclusions:
This project has demonstrated the application of systematic uncertainty analysis methods for photochemical air quality models. For the CIT trajectory model applied to receptor sites in the South Coast Air Basin on August 27-28, 1987, our research indicates that uncertainties in predicted ozone concentrations range from about 20 to 60%, depending on the trajectory endpoint. As shown by Bergin and Milford (1999) good agreement between model results and spatially interpolated ozone concentrations suggests that the uncertainties for this application may be somewhat smaller than estimated without regard to model performance. The uncertainty estimates and major sources of uncertainty found in this study are specific to the model, simulation conditions, observations and error estimates that we used. In particular, compared to many other urban areas, the South Coast Air Basin is unique in having a relatively small impact from transported pollutants and having a relatively dense network of air quality and meteorological monitoring stations to provide model inputs. Although the quantitative results are specific to the case studied, the techniques demonstrated in this research are generalizable to other air quality modeling applications. Significant new methods developed in this study include the use of fuel sales and remote sensing data for estimating motor vehicle emissions and associated uncertainties; objective techniques for estimating wind field uncertainties; and the application of Bayesian Monte Carlo analysis to photochemical air quality models. The results of such analyses can help decision makers gauge the reliability of models used for assessing the effect of proposed control measures and help set priorities for further research to improve predictive capabilities.Journal Articles on this Report : 2 Displayed | Download in RIS Format
Other project views: | All 6 publications | 2 publications in selected types | All 2 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Bergin MS, Noblet GS, Petrini K, Dhieux JR, Milford JB, Harley RA. Formal uncertainty analysis of a Lagrangian photochemical air pollution model. Environmental Science and Technology 1999;33(7):1116-1126. |
R824792 (Final) |
Exit |
|
Bergin MS, Milford JB. Application of Bayesian Monte Carlo analysis to a Lagrangian photochemical air quality model. Atmospheric Environment 2000;34(5):781-792. |
R824792 (Final) |
Exit Exit Exit |
Supplemental Keywords:
air, ambient air, atmosphere, ozone, particulates, oxidants, modeling, central, Colorado, CO, Region 8, RFA, Scientific Discipline, Air, particulate matter, Air Quality, mobile sources, Environmental Monitoring, tropospheric ozone, Atmospheric Sciences, Ecological Risk Assessment, Ecology and Ecosystems, ambient air quality, particulates, ozone occurrence, air quality models, ambient measurement methods, ozone, ambient air, National Ambient Air Quality Standard, Los Angeles, control measure modeling, modeling studies, atmospheric chemistryRelevant Websites:
Web site for Jana Milford: http://spot.colorado.edu/~milford Exit ExitProgress 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.