Simulating Pollutant Dispersion from Accidental Fires with a Focus on Source Characterization

Background. Storage tanks in oil and gas processing facilities contain large volumes of flammable compounds. Once the fuel-air mixture is ignited, it may break out into a large fire or explosion. The growing interest in monitoring air quality and assessing health risks makes the evaluation of the consequences of a fire an important issue. Atmospheric dispersion models, which allow for simulation of the spatial distribution of pollutants, represent an increasingly widespread tool for this type of evaluations. Objectives. The present study discusses the set up and results of a modeling study relevant to a hypothesized fire in an oil refinery. Methods. After choosing the most suitable dispersion models, i.e. the Lagrangian model SPRAY and the puff model CALPUFF, estimation of the required input data is discussed, focusing on the source variables, which represent the most uncertain input data. The results of the simulations were compared to regulatory limits to effectively evaluate the environmental consequences. Finally, a sensitivity analysis was employed to identify the most influential variables. Results. The simulation results revealed that ground concentration values were far below the cited long-term limits. However, the most interesting outcome is that depending on the dispersion model and the source type modeled, different results may be obtained. In addition, the sensitivity study indicates that the source area is the most critical variable, since it determines a significantly different behavior depending on the modeled source types, producing, in some cases, variability in the pollutant ground concentrations on selected receptors up to +/- 60%. Conclusions. Depending on the selected model and the algorithms available to describe the physics of emission, the results showed a different sensitivity to the input variables. Although this can be explained from a mathematical point of view, the problem remains of choosing case by case the option that best approximates the real behavior of the incidental source under investigation.