Sensitivity Analysis of a Large-Scale Air Pollution Model by Using Effective Stochastic Approaches

This paper aims an exploration of model output sensitivity in the air pollution transport. A comprehensive experimental study of highly efficient stochastic approaches based on the Van der Corput sequence and its modification with different bases for multidimensional integration has been done. A comparison with the Latin Hypercube Sampling (LHS) and Fibonacci based lattice rule (FIBO) is given. The algorithms have been successfully applied to compute global Sobol sensitivity measures corresponding to the six chemical reactions rates and four different groups of pollutants. The numerical tests show that the stochastic algorithms under consideration are efficient for the multidimensional integrals and especially for computing small by value sensitivity indices.