THE GEOGRAPHIC AND STATISTICAL ANALYSIS OF AIR QUALITY DATA IN THE UNITED STATES (OHIO)

This dissertation contains the development and the application of analytic procedures for examining and exploring some air quality data collected by the Environmental Protection Agency from 1974 through 1976. They are collected at monitoring stations most of which are in metropolitan areas. These data are irregularly distributed discrete point measurements. The techniques explored here may be useful in other disciplines with the same type of data.; The analysis is concentrated on two pollutants, suspended particulate and sulfur dioxide. There are two reasons for this restriction: (i) they are the most heavily monitored and (ii) they are of interest to the health field. The state of Ohio is utilized as an example in most of these analyses. This is because Ohio is the most thoroughly monitored state in the United States. A list of the limitations of these data is given.; Interpolation schemes are explored and a model is chosen which is a two-dimensional analogue of the moving average model in time series. The model is; (DIAGRAM, TABLE OR GRAPHIC OMITTED...PLEASE SEE DAI); where, e(,i) = the estimated value at a point i; x(,j) = a measured value at point j; d(,ij) = the distance from the data point to the point of estimation; d(,0) = the smoothing parameter. The choice of d(,0) has been explored in great detail. Cross-validation was used and several measures for the "best" d(,0) were examined. This led to the development of a much more efficient method for choosing a smoothing parameter, the concept of local variability as a function of disk radius. Each disk radius corresponds to a d(,0), so by minimizing the local variability function the most appropriate d(,0) can be chosen. Local variability functions were calculated for Ohio, New York and Florida. This analysis as opposed to cross-validation makes the task of modeling the entire United States a much smaller one. This model combined with cross-validation has been useful in detecting outliers in these data.; The evaluation of the moving average model led to comparing to Akima's method of bivariate linear interpolation. A cross-validatory comparison for adequacy of estimation was done. Also, contour maps using each method are drawn and compared. The local variability function analysis allows for comparison by cross-validation to not be a two-deep cross-validatory choice. Some drawbacks to comparing cross-validation estimates are pointed out. How different goals may prescribe different estimation techniques is discussed.; The potential for further research in this field is shown. Time, which may be important in these analyses, has not been included because of data availability limitations. Using a time parameter similar to d(,0), the current distance parameter, has been suggested. Simulations may also be useful in evaluating the moving average model. The distributional theory of the local variability theory function is yet to be explored.