"INTRODUCTION Background Since the passage of the Clean Air Act@ the Environmental Protection Agency (EPA) has used numerous air quality models that relate source emissions and meteorological data to the expected air quality. The EPA has relied upon the predictive capability of air quality models to identify the levels of control required to solve industrial and urban air pollution problems. The models are mathematical representations of the complex physical and chemical processes involved in the dispersion@ transformation and deposition of pollutants. However@ the complexity and turbulent stochastic properties of the atmosphere limit the predictive capability of even the best models. Uncertainties in much of the input data further limit predictive capability. Through a series of guideline documents@ 1@2@3 EPA has attempted to standardize modeling approaches for regulatory applications. The desire for consistency in applying models has led EPA to recommend a particular model for each of a series of general applications depending upon source type@ land use@ topography@ pollutants and averaging times of interest. Most of the models selected by the EPA are based on the steady-state Gaussian point-source formulae developed in the 1940's with numerous enhancements for transformation@ deposition@ decay@ buildup@ depletion@ and multiple source and receptor configurations. These Gaussian models utilize empirical parameters derived from field experiments to predict dispersion behavior as a function of meteorological conditions. A number of tracer dispersion programs conducted during the 1950s and 1960s were used to estimate these dispersion parameters. Data from these tracer experiments were analyzed prior to wide-scale use of computers and of advanced numerical analysis routines. In developing empirical dispersion parameters@ substantial uncertainty was recognized and tolerated. In 1979@ TRC Environmental Consultants@ Inc. (TRC) undertook a study for the American Petroleum Institute (API) to evaluate the performance of current air quality models@ using data from tracer dispersion programs. This study found frequent large discrepancies between observed and predicted tracer concentrations. Lloyd Hellums of Phillips Petroleum concluded that improved dispersion parameters might result from the re-analysis of these tracer data with modem mathematical fitting techniques. He therefore developed a series of computer programs to analyze tracer experiments@ using a non-linear least-squares optimization routine proprietary to Phillips Petroleum. The method is called the Single Source Mass Transfer (SSMT) method. Hellums has now analyzed data from twelve tracer programs to estimate dispersion coefficients for each individual tracer experiment. API contracted TRC to compare the results obtained by Hellums with previous dispersion parameter estimates and to evaluate the potential for utilizing these results to develop improved air quality models. Objectives and Approach The two primary objectives of the project are to estimate the uncertainties inherent in Gaussian model predictions and to assess the feasibility of improving model performance through the use of revised dispersion parameters. The SSMT uses a non-linear least-squares optimization technique to derive best-fit estimates of dispersion parameters. The parameters are chosen to minimize the differences between observed tracer concentrations and Gaussian model predictions. The differences (often termed ""residuals"") which remain@ following optimization@ generally correspond to observed concentration values which do not follow a Gaussian distribution. These residual values set a lower limit on model uncertainty: a ""perfect"" steady-state Gaussian air quality model cannot be expected to provide a better match to observed concentrations than the optimized solution for each experiment. In addition to non-Gaussian distributions@ several other issues must be confronted in order to evaluate the feasibility of improving air quality models via this optimization approach. These include the degree of difference between current and proposed dispersion parameters@ the magnitude of inherent uncertainties@ and the relationship between observed dispersion and meteorological conditions. For (his study@ the major criterion used in investigating the potential for model improvement was that the SSMT dispersion parameters should provide predicted concentration values which consistently match observed values. The optimization method operates to minimize residual values and routinely provides ""goodness of fit"" measures for each SSMT solution. Hellums incorporated three alternative fitting schemes into the SSMT programs and compared results from each scheme. TRC also assessed the relative merits of these three alternatives. From these assessments@ one SSMT option was selected for more detailed analysis. From a practical viewpoint@ dispersion parameters from individual tracer experiments cannot be used to predict dispersion unless they are incorporated within a model framework keyed to meteorological@ site@ and source conditions. The general model development approach attempts to combine estimates of dispersion from experiments with similar types of key conditions through an ""atmospheric stability class'' that defines an index of atmospheric turbulence. Several stability classification schemes are used currently by the technical and regulatory community. Prediction schemes in existing air quality models utilize the atmospheric stability class to select appropriate empirical dispersion parameters for a particular event. Additional steps in assessing the SSMT results as a basis for improved models included a comparison with earlier estimates of dispersion parameters and examination of the SSMT dispersion coefficients as a function of several standard classification schemes During the course of an earlier study for API@4 TRC developed dispersion estimates for a number of tracer experiments A relatively simple calculation scheme was used@ similar to the techniques from which current model parameters were derived SSMT results have been compared to these simple estimates and to the generalized dispersion algorithms employed in existing models A comparison of SSMT dispersion coefficients as a function of the stability class was undertaken to determine whether current methods of stability classification arc adequate In the following sections of this report@ our evaluation of the SSMT method is presented Section 2 defines the SSMT method and describes the three alternative optimization schemes In Section 3@ a description of the tracer experiment data base is given In Section 4@ results for several selected experiments are presented to illustrate the method Sections 5 and 6 present statistical summaries comparing SSMT results with observed concentration values and with other dispersion parameter estimates Section 7 examines SSMT dispersion parameters as a function of atmospheric stability class Project findings and conclusions are presented in Section 8"
API PUBL 4376-1983 history
1983API PUBL 4376-1983 Development of Improved Dispersion Estimates in Tracer Experiments