Modeling the spatial behavior of the meteorological drivers' effects on extreme ozone

At its most extreme levels, ground-level ozone is most harmful to human health, and meteorological conditions play a critical role in such episodes. In this work, our aim is to better understand how the primary meteorological drivers' effects on extreme ozone vary over the Southeast and Mid-Atlantic region of the USA. We employ a model based on a bivariate extreme value framework that finds the linear combination of a set of meteorological covariates that has the strongest tail dependence with ground-level ozone. In order to gain knowledge about the spatial behavior of the meteorological drivers, we spatially model the coefficients, which relate the covariates to extreme ozone. Because inference for our extreme value model is not likelihood based, we utilize a two-stage modeling procedure: first, estimating the coefficients in our extreme value model and their associated uncertainties and then using these to fit a multivariate spatial process. We analyze data from 160 air quality stations located in the Environmental Protection Agency Regions 3 and 4, producing estimated spatial surfaces via the fitted spatial process and co-kriging. We find that the relative contribution of the driving meteorological variables to extreme ozone differs between the northern and southern portions of the study region. For instance, we find that air temperature is more important in the northern portion of the region, while low humidity is more influential in the southern portion of the region.