Sensitivity Analysis of a Large-Scale Air Pollution Model: Numerical Aspects and a Highly Parallel Implementation

The Unified Danish Eulerian Model (UNI-DEM) is a powerful air pollution model, used to calculate the concentrations of various dangerous pollutants and other chemical species over a large geographical region (mostly Europe). It takes into account the main physical and chemical processes between these species, the emissions, the deposition, advection and diffusion in dependence with the changing meteorological conditions. This large and complex task is split into submodels responsible for the main physical and chemical processes. In the chemical submodel there is a number of parameters, responsible for the speed of the corresponding chemical reactions. By simultaneous variation of these parameters we produced a set of multidimensional discrete functions. These are necessary for variance-based sensitivity analysis by using adaptive Monte Carlo approaches, which is subject to another paper. These huge computational tasks require extensive resources of storage and CPU time. A highly parallel implementation of the UNI-DEM has been created for this purpose and implemented on the new IBM BlueGene/P supercomputer, the most powerful parallel machine ever in Bulgaria. Some details of this implementation and a set of results obtained by it are presented in this paper.