Errors in CMAQ Prediction of Ozone Concentration

Global scale pollution and climate change rise among the most controversial topics in society today. The decisions of environmental inspectors, healthcare reformers, scientists, and policy makers rely heavily on the results of statistical pollution models derived by the scientists. Deterministic atmospheric chemistry models appear among the most useful and accurate tools in air quality decision making. Our goal is to utilize non-spatial (stepwise selection) and spatial analysis (Kriging) to make improvement on ozone concentration predictions.  We accessed and modified a large-scale dataset containing the observed measurements and mathematical predictions from “Community Multi-scale Air Quality” (CMAQ) on weather conditions in 85 locations. By selecting variables that contribute the most to errors, we generated a linear model using these variables to model the errors generated by CMAQ predictions. Our computational results generated by statistical software R showed that 9 out of 20 variables have near-zero p-values (<2e-16) as well as significant intercepts beta values (-2.02848 to 6.68224).  We concluded that error-inducing variables are same from sites to sites, although some are most spatially important than others. Our list of these variables is temperature, relative humidity, precipitation, wind speed, NOx chemical group, etc. By identifying these variables, it would allow the EPA and other CMAQ users to more accurately predict ozone levels throughout the country.