Interannual variation in meteorologically adjusted ozone levels in the eastern United States: A comparison of two approaches

Assessing the influence of abatement efforts and other human activities on ozone levels is complicated by the atmosphere's changeable nature. Two statistical methods, the dynamic linear model (DLM) and the generalized additive model (GAM), are used to estimate ozone trends in the eastern United States and to adjust for meteorological effects. The techniques and resulting estimates are compared and contrasted for four monitoring locations chosen through principal components analysis to represent regional patterns of ozone concentrations. After adjustment for meteorological influence, overall downward trends are evident at all four locations from 1997 to 2004. The results indicate that the two methods’ estimates of ozone changes agree well. When such estimates are needed quickly, or when many similar, but separate analyses are required, the ease of implementation and relative simplicity of the GAMs are attractive. The DLMs are much more flexible, readily addressing such issues as autocorrelation, the presence of missing values, and estimation of long-term trends or cyclical patterns. Implementation of DLMs, however, is typically more difficult, and especially in the absence of an experienced practitioner, they may be better reserved for in-depth analyses.