Historical trends and ozone forecasting for urban regions of South Texas using statistical and heuristic techniques

A comprehensive overview of the historical trends of ozone air quality in major urban centers of South Texas is provided in this study. The historical data was then utilized to develop a forecasting model using statistical and heuristic techniques.; A variety of robust nonparametric statistical methods was used to analyze raw trends incorporating both the one-hour and eight-hour standards. The results showed that generally there was a decreasing trend in ozone concentrations in the study sites for both the one and eight hour standards. However, in some cases there was a trend reversal which showed increases when the more stringent eight-hour standard was applied.; Trends are also derived by the use of the Kolmogorov-Zurbenko filter, a low-pass moving average technique for removing background noise and making meteorological adjustments that discern anthropogenic emissions to air quality. The Kolmogorov-Zurbenko filter is usually coupled with the statistical linear regression technique to adjust for emissions to air quality. This research revealed about 10% of ozone variation over the study period using this technique.; A third trend analysis highlighted in the study combined the Kolmogorov-Zurbenko technique, but incorporating artificial neural networks technique in lieu of the statistical linear regression technique produced results showing decreasing trends at four sites in the study which explained 10.38% to 28.76% variability.; This study presents a holistic synopsis of usage of current-state-of-science heuristic models in order to forecast daily maximum one-hour and eight-hour ozone concentrations in the semi-arid South Texas region. The two models used in this study are the Statistical Analysis Linear Regression (LR) model, and the Artificial Neural Networks multilayer perceptron (MLP) type model which have been developed and applied at four ozone observational sites: CAMS04 and CAMS21 in the Corpus Christi area, CAMS23 in San Antonio, and CAMS80 in Victoria. Ideally, the models should be developed using an optimum quantity of data and input parameters. Findings from this study indicate that only two years of data would be sufficient to develop a model instead of using large quantities of data. The analyses of trends in observed data is important to understand the ozone formation and identify the periods of high ozone concentrations in a region.; Statistical Regression Analysis and Artificial Neural Networks models were developed to forecast high ozone episodes of Corpus Christi, San Antonio and Victoria. The predictive capabilities of these models were compared to determine their accuracy in forecasting daily maximum eight-hour ozone and peak one-hour ozone concentration by different model evaluation statistics. Results of the study indicated that both models were able to capture trends in ozone time series, but they were unable to accurately forecast peak ozone values.; Between the two models, the performance of the Statistical Analysis model LR - RMSE averages ranged from 13.55 to 15.96 and the Artificial Neural Networks model MLP - RMSE averages ranged from 12.56 to 14.96, there seems to be very little difference in the forecasting capabilities. Also, it should be noted that according to forecasting skill proficiency, the 8-hour ozone standard provided the highest probability of detection of a high ozone forecast than did the 1-hour standard.; There is still a degree of uncertainty associated with the result of the findings concerning the models ability to make accurate forecast. Previous studies have also noted that the difference in the performance superiority of one modeling technique over the other is suspect. Where the skill proficiency of one model fails the other may succeed and vice versa.