Results of an Extensive Evaluation of the Kinematic Simulation Particle Model Using Tracer and Wind Tunnel Experiments

The Kinematic-Simulation-Particle (KSP) model is a new type of atmospheric dispersion model having the intrinsic capacity to predict second-by-second concentration patterns and fluctuations that can be critical to the assessment of odor and hazardous chemical exposure problems. KSP uses the mean flow fields and a realistic spectrum of turbulent eddies to transport tracer particles. Just as in nature, the instantaneous distribution of these tracer particles yields concentration field 'snapshots' that may be analyzed to yield critical statistics (e.g., peaks, durations above thresholds, higher moments) as well as the more usual, longer time averages and ensemble means. KSP represents a true 'parallel universe' model of the atmosphere and can be used to investigate phenomena such as relative- versus absolute-diffusion as well as more applied problems. In addition, as the KSP model can be driven by the meteorological model CALMET, it has the capability to assess the consequences of air pollution emissions over a broad range of: spatial and temporal scales; terrain and land use environments; and non-steady meteorological and release conditions.

The KSP system has recently undergone an extensive evaluation process, funded by the German EPA (the Umweltbundesamt), to determine biases and quantify model uncertainties. The present paper examines some of the key model enhancements (e.g., convective towers, eddy transport by larger eddies) and summarizes the results of comparisons with data obtained from several full-scale tracer experiments and wind tunnel studies of concentration fluctuations. This evaluation has shown that the current version of KSP:

* produces a range of ensemble -average concentration estimates that generally encompass the predictions of simple Gaussian plume models and Lagrangian particle models.

* performs similarly to simpler models in matching the crosswind-integrated and arc-maximum concentrations obtained during several tracer campaigns. This is true both for cases where most models perform well (e.g., Kincaid, Copenhagen, Prairie Grass arcs at 50m-200m) and also for cases where models generally do poorly (e.g., Lillestroem, Prairie Grass 400m-800m arcs).

* predicts a variability in concentration estimates on the order of a factor-of-two or more for tracer exposure durations of 10-60 minutes. That is, s_c/C is of order unity.

* generates cumulative frequency distributions (CFDs) of concentrations that have the same qualitative behavior as those measured in the wind tunnel. These modeled CFDs are very sensitive to the assumed turbulence power spectra and its profile, nS(n,z), and assumed spectral component time scales.