(81bu) Sensitivity ANALYSIS of Atmospheric Dispersion MODELING IN Emergency Situation

Each day, accidents or pre-accidents involving hazardous materials or chemical processes are managed by Emergency Services. The Bureau for Analysis of Industrial Risks and Pollution (BARPI in FRANCE) highlighted the fact that approximately 10% of these accidents involve air pollution. For each phase of an emergency situation (threat, accident, post-accident), in order to bring a best adapted operational answer, it is necessary to determine with accuracy the concentrations of gas at which people will be exposed. Moreover, in emergency situation, modelling is based on vague, incomplete or missing information.

An analysis of sensitivity was carried out in order to determine the most influential parameters on the estimation of safety distances. The goal is to identify the parameters for which it is necessary to pay a very detailed attention, either in obtaining a value, or in the choice of a default value, or in its treatment. The first step of the study was to define the system to be analyzed. The second step consisted in working out a complete model of evaluation of safety distances in emergency situation. This was done by combining an existing dispersion model (SLAB) with a specifically developed application for the calculation of the input parameters of the dispersion model. The second objective of this model was also to automate calculations for the making of the sensitivity analysis described in the present paper. The input data of the model are operational data which can be collected from the accident place and the output data are safety distances corresponding to effect thresholds. Keeping in mind that this model must be used in emergency situation, it was tried to respect three principles: minimize the number of input data, select methods and models having fast computing times but describing as accurately as possible each physical phenomenon. The third step consisted in selecting test cases representative of an accidental situation involving hazardous chemicals. Two accident scenarios were studied: leak on a wagon of ammonia (toxic liquefied gas) and leak on a wagon of propane (flammable liquefied gas). After that, it was possible to characterize the range of possible or probable values of the whole input parameters needed to make calculation. A fourth step consisted in applying to this system three methods of sensitivity analysis: a screening method (Morris method), a local sensitivity analysis and a method based on the experimental designs known as Taguchi method. Finally, the application of these three methods, on the same case of study, made it possible to highlight the points of convergence / divergence of the methods, their advantages / disadvantages in the objective to identify the parameters mostly responsible for the uncertainty of prediction. The study of sensitivity also made it possible to identify the accident configurations for which a conservative deterministic approach is possible. Finally, the parameters for which it is useless / useful to engage considerable efforts to recover a reliable value were highlighted.