Atmospheric dispersion is highly variable especially in a convective boundary layer (CBL) with light winds. Such variability occurs in the plume height, dispersion parameters, and surface concentrations. For an elevated source, the root-mean-square (rms) fluctuating concentration observed at short distances downstream (< 5 km) is typically as large as the mean concentration. Knowledge of the variability is important for estimating maximum ground-level concentrations, predicting odors, and evaluating dispersion models.

This paper focuses on the statistical variability of plume dispersion using a Lagrangian particle dispersion model (LPDM) driven by large-eddy simulations (LESs) of the CBL. The coupled modeling approach was used previously to simulate the mean concentration fields in the CBL (Weil et al., 2004) and found to agree well with laboratory and field data. In this paper, we extend the approach to the variability of plume properties---the plume height, dispersion parameters, concentrations, etc. For each source, 30 realizations of the concentration field, each approximating a 1/2-h average, are obtained to produce an ensemble for the same average LES flow conditions, mean wind, surface height flux, CBL height, etc. Ensembles of 30 realizations are developed for each of three source heights---surface, near-surface, and elevated---and for each of them we find the overall statistical properties.

We compare the LPDM results with field observations of plume properties from two experiments---CONDORS (Briggs, 1993) for surface and elevated sources and Prairie Grass (Barad, 1958) for a surface source. The overall conclusion is that the LPDM-LES model generates a realistic range of dispersion realizations and rms values that match observations, while also producing mean values in agreement with the observations. This holds for the plume height or trajectory, vertical dispersion, and surface concentrations. Results generated by the LPDM-LES approach coupled with those from observations and experiments should be useful in evaluating simpler dispersion models.