An Examination of the Performance of Models for Dispersion in the Convective Boundary Layer

Abstract

The past few years have seen the development of several models for dispersion of pollutants released from tall stacks into the daytime convective boundary layer. These models (Weil and Brower, 1984; Venkatram and Vet, 1981; Hanna and Paine, 1987) incorporate the latest understanding of the turbulent structure of the convective boundary layer. In general, they perform better than models based on empirical stability classification systems in explaining observations of ground-level concentrations. Although this improvement does represent a substantial increase in the explained variance of the observations, the actual value of the explained variance, as represented by the coefficient of determination r², is relatively small. For example, the new model developed by Weil et al. (1987) explains only 34 percent of the observed variance. Hanna and Paine (1987) acknowledge the fact that the correlation between observations and predictions from their model is not significantly different from zero. Although the geometric mean of the ratio of the observed to the predicted concentrations is close to unity, the expected deviation between model predictions and observations is large. Weil et al (1987) find that the geometric standard deviation of the ratio of the predicted to observed concentrations is around two (2). Assuming a lognormal distribution of observed concentrations about the predicted mean, the standard deviation of 2 implies that 95 percent of the observations lie only within a factor of four of a given model prediction.