Statistical models for distributions of ambient fine particulate matter

This study evaluates the usefulness of structured non-linear regression models for the prediction of annual ambient fine particulate matter (FPM) concentration distributions. The method developed in this study provides a way to examine and display results for the yearly distribution of FPM when testing emissions control strategy performance. The models are developed using three daily gaseous pollutant concentrations (oxides of nitrogen (NO,), sulfur dioxide (SO2), and total hydrocarbons (THC)) and four meteorological measures (wind speed, temperature, relative humidity and precipitation) as explanatory variables. The models are fitted using data from the North Long Beach, Rubidoux (Riverside) and Azusa stations in Los Angeles County and Riverside County, CA for a recent 7-year period (1988-1994). The statistical model is tested for the year 1995 based on precursor concentrations and meteorological conditions in that year, and found to provide reasonably good prediction, though the annual average FPM concentration is overestimated by an average of 26 percent across the three stations. The response surfaces Of PM2.5 concentrations with respect to all input variables are plotted, and the predicted changes in daily, annual average and annual 98th percentile base-year (1995) PM2.5 concentrations are predicted for different precursor reductions. The predicted effects of precursor reductions are further explored by comparing predicted and observed FPM concentrations for 1999 (though the absence of THC data for this year restricts this comparison to plausible ranges). The method developed in this study provides a way to examine and display results for the predicted concentration distributions when evaluating emission control strategy performance. The potential usefulness and limitations of a statistical model of this type are discussed.