Sensitivity Analysis of an Air Pollution Model by Using Quasi-Monte Carlo Algorithms for Multidimensional Numerical Integration

Sensitivity analysis is a powerful tool for studying and improving the reliability of large and complicated mathematical models. Air pollution and meteorological models are in front places among the examples of such models, with a lot of natural uncertainties in their input data sets and parameters. We present here some results of our global sensitivity study of the Unified Danish Eulerian Model (UNI-DEM). One of the most attractive features of UNI-DEM is its advanced chemical scheme - the Condensed CBM IV, which consider in detail a large number of chemical species and numerous reactions between them. Four efficient stochastic algorithms (Sobol QMC, Halton QMC, Fibonacci lattice rule and Latin hypercube sampling) have been used and compared by their accuracy in studying the sensitivity of ammonia and ozone concentration results with respect to the emission levels and some chemical reactions rates. The numerical experiments show that the stochastic algorithms under consideration are quite efficient for this purpose, especially for evaluating the contribution of small by value sensitivity indices.