Sensitivity Analysis of an Air Pollution Model with Using Innovative Monte Carlo Methods in Calculating Multidimensional Integrals

Large-scale models are mathematical models with a lot of natural uncertainties in their input data sets and parameters. Sensitivity analysis (SA) is a powerful tool for studying the impact of these uncertainties on the output results and helps to improve the reliability of these models. In this article we present some results of a global sensitivity study of the Unified Danish Eulerian Model (UNI-DEM). A large number of heavy numerical experiments must be carried out in order to collect the necessary data for such comprehensive sensitivity study. One of the largest supercomputers in Europe and the most powerful in Bulgaria, the petascale EuroHPC supercomputer Discoverer is used to perform efficiently this huge amount of computations. One of the most important features of UNI-DEM is its advanced chemical scheme, called Condensed CBM IV, which considers a large number of chemical species and all significant reactions between them. The ozone is one of the most harmful pollutants, that is why it is important for many practical applications to study it precisely. Stochastic methods based on Adaptive approach and Sobol sequences are used for computing the corresponding sensitivity measures. We show by experiments that the stochastic algorithms for calculating the multidimensional integrals under consideration are one of the best stochastic techniques for computing the small in value sensitivity indices.