Examining the Observed and Modeled Sensitivities of Air-quality Extremes to Meteorological Drivers using Advanced Statistical Techniques

Abstract:

Changes in local climatology are likely driving changes in local air-quality, both in terms of mean pollutant levels as well as the frequencies and magnitudes of extreme events. While the impact of changing meteorology on O₃ and PM_2.5 has been explored previously, much of the existing literature has focused on averages and linear regressions, statistical techniques that largely ignore extreme behavior and tail dependence. Since current air-quality standards often include limits on high quantiles of pollutant level observations (e.g. annual 4^th highest daily maximum 8-hour concentration of O₃ and 98^th percentile of PM_2.5 in the United States), statistical analyses that do not focus on tail dependence will be unable to fully evaluate impacts on exceedance frequencies. Using methodologies based on quantile regression (QR) and extreme value theory (EVT), tools specifically developed for the analysis of heavy-tailed phenomena, we analyze relationships between meteorology and extreme pollution episodes in the United States, both in the observed data record and in modeled output generated by the Community Earth System Model (CESM). Through this analysis, we propose a statistical framework for the identification of the meteorological drivers of these air-quality extremes, the evaluation of modeled extremes, and the improvement of future extreme projections based on observed sensitivities and assumed climatological changes.